?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
irqchip/zevio: Use irq_data_get_chip_type() helper
[linux/fpc-iii.git] / fs / btrfs / inode.c
bloba70c5790f8f5908f08f606d097fe33e6966af36b
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
52 #include "xattr.h"
53 #include "tree-log.h"
54 #include "volumes.h"
55 #include "compression.h"
56 #include "locking.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
59 #include "backref.h"
60 #include "hash.h"
61 #include "props.h"
62 #include "qgroup.h"
63
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
67 };
68
69 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_dir_ro_inode_operations;
72 static const struct inode_operations btrfs_special_inode_operations;
73 static const struct inode_operations btrfs_file_inode_operations;
74 static const struct address_space_operations btrfs_aops;
75 static const struct address_space_operations btrfs_symlink_aops;
76 static const struct file_operations btrfs_dir_file_operations;
77 static struct extent_io_ops btrfs_extent_io_ops;
78
79 static struct kmem_cache *btrfs_inode_cachep;
80 static struct kmem_cache *btrfs_delalloc_work_cachep;
81 struct kmem_cache *btrfs_trans_handle_cachep;
82 struct kmem_cache *btrfs_transaction_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
85
86 #define S_SHIFT 12
87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
88 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
89 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
90 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
91 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
92 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
93 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
94 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
95 };
96
97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
98 static int btrfs_truncate(struct inode *inode);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
100 static noinline int cow_file_range(struct inode *inode,
101 struct page *locked_page,
102 u64 start, u64 end, int *page_started,
103 unsigned long *nr_written, int unlock);
104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
105 u64 len, u64 orig_start,
106 u64 block_start, u64 block_len,
107 u64 orig_block_len, u64 ram_bytes,
108 int type);
109
110 static int btrfs_dirty_inode(struct inode *inode);
111
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode *inode)
114 {
115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
116 }
117 #endif
118
119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
120 struct inode *inode, struct inode *dir,
121 const struct qstr *qstr)
122 {
123 int err;
124
125 err = btrfs_init_acl(trans, inode, dir);
126 if (!err)
127 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
128 return err;
129 }
130
131 /*
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
135 */
136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_path *path, int extent_inserted,
138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
140 int compress_type,
141 struct page **compressed_pages)
142 {
143 struct extent_buffer *leaf;
144 struct page *page = NULL;
145 char *kaddr;
146 unsigned long ptr;
147 struct btrfs_file_extent_item *ei;
148 int err = 0;
149 int ret;
150 size_t cur_size = size;
151 unsigned long offset;
152
153 if (compressed_size && compressed_pages)
154 cur_size = compressed_size;
155
156 inode_add_bytes(inode, size);
157
158 if (!extent_inserted) {
159 struct btrfs_key key;
160 size_t datasize;
161
162 key.objectid = btrfs_ino(inode);
163 key.offset = start;
164 key.type = BTRFS_EXTENT_DATA_KEY;
165
166 datasize = btrfs_file_extent_calc_inline_size(cur_size);
167 path->leave_spinning = 1;
168 ret = btrfs_insert_empty_item(trans, root, path, &key,
169 datasize);
170 if (ret) {
171 err = ret;
172 goto fail;
173 }
174 }
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
178 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
179 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
180 btrfs_set_file_extent_encryption(leaf, ei, 0);
181 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
182 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
183 ptr = btrfs_file_extent_inline_start(ei);
184
185 if (compress_type != BTRFS_COMPRESS_NONE) {
186 struct page *cpage;
187 int i = 0;
188 while (compressed_size > 0) {
189 cpage = compressed_pages[i];
190 cur_size = min_t(unsigned long, compressed_size,
191 PAGE_CACHE_SIZE);
192
193 kaddr = kmap_atomic(cpage);
194 write_extent_buffer(leaf, kaddr, ptr, cur_size);
195 kunmap_atomic(kaddr);
196
197 i++;
198 ptr += cur_size;
199 compressed_size -= cur_size;
200 }
201 btrfs_set_file_extent_compression(leaf, ei,
202 compress_type);
203 } else {
204 page = find_get_page(inode->i_mapping,
205 start >> PAGE_CACHE_SHIFT);
206 btrfs_set_file_extent_compression(leaf, ei, 0);
207 kaddr = kmap_atomic(page);
208 offset = start & (PAGE_CACHE_SIZE - 1);
209 write_extent_buffer(leaf, kaddr + offset, ptr, size);
210 kunmap_atomic(kaddr);
211 page_cache_release(page);
212 }
213 btrfs_mark_buffer_dirty(leaf);
214 btrfs_release_path(path);
215
216 /*
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
220 *
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
224 */
225 BTRFS_I(inode)->disk_i_size = inode->i_size;
226 ret = btrfs_update_inode(trans, root, inode);
227
228 return ret;
229 fail:
230 return err;
231 }
232
233
234 /*
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
238 */
239 static noinline int cow_file_range_inline(struct btrfs_root *root,
240 struct inode *inode, u64 start,
241 u64 end, size_t compressed_size,
242 int compress_type,
243 struct page **compressed_pages)
244 {
245 struct btrfs_trans_handle *trans;
246 u64 isize = i_size_read(inode);
247 u64 actual_end = min(end + 1, isize);
248 u64 inline_len = actual_end - start;
249 u64 aligned_end = ALIGN(end, root->sectorsize);
250 u64 data_len = inline_len;
251 int ret;
252 struct btrfs_path *path;
253 int extent_inserted = 0;
254 u32 extent_item_size;
255
256 if (compressed_size)
257 data_len = compressed_size;
258
259 if (start > 0 ||
260 actual_end > PAGE_CACHE_SIZE ||
261 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
262 (!compressed_size &&
263 (actual_end & (root->sectorsize - 1)) == 0) ||
264 end + 1 < isize ||
265 data_len > root->fs_info->max_inline) {
266 return 1;
267 }
268
269 path = btrfs_alloc_path();
270 if (!path)
271 return -ENOMEM;
272
273 trans = btrfs_join_transaction(root);
274 if (IS_ERR(trans)) {
275 btrfs_free_path(path);
276 return PTR_ERR(trans);
277 }
278 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
279
280 if (compressed_size && compressed_pages)
281 extent_item_size = btrfs_file_extent_calc_inline_size(
282 compressed_size);
283 else
284 extent_item_size = btrfs_file_extent_calc_inline_size(
285 inline_len);
286
287 ret = __btrfs_drop_extents(trans, root, inode, path,
288 start, aligned_end, NULL,
289 1, 1, extent_item_size, &extent_inserted);
290 if (ret) {
291 btrfs_abort_transaction(trans, root, ret);
292 goto out;
293 }
294
295 if (isize > actual_end)
296 inline_len = min_t(u64, isize, actual_end);
297 ret = insert_inline_extent(trans, path, extent_inserted,
298 root, inode, start,
299 inline_len, compressed_size,
300 compress_type, compressed_pages);
301 if (ret && ret != -ENOSPC) {
302 btrfs_abort_transaction(trans, root, ret);
303 goto out;
304 } else if (ret == -ENOSPC) {
305 ret = 1;
306 goto out;
307 }
308
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
310 btrfs_delalloc_release_metadata(inode, end + 1 - start);
311 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
312 out:
313 /*
314 * Don't forget to free the reserved space, as for inlined extent
315 * it won't count as data extent, free them directly here.
316 * And at reserve time, it's always aligned to page size, so
317 * just free one page here.
318 */
319 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
320 btrfs_free_path(path);
321 btrfs_end_transaction(trans, root);
322 return ret;
323 }
324
325 struct async_extent {
326 u64 start;
327 u64 ram_size;
328 u64 compressed_size;
329 struct page **pages;
330 unsigned long nr_pages;
331 int compress_type;
332 struct list_head list;
333 };
334
335 struct async_cow {
336 struct inode *inode;
337 struct btrfs_root *root;
338 struct page *locked_page;
339 u64 start;
340 u64 end;
341 struct list_head extents;
342 struct btrfs_work work;
343 };
344
345 static noinline int add_async_extent(struct async_cow *cow,
346 u64 start, u64 ram_size,
347 u64 compressed_size,
348 struct page **pages,
349 unsigned long nr_pages,
350 int compress_type)
351 {
352 struct async_extent *async_extent;
353
354 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
355 BUG_ON(!async_extent); /* -ENOMEM */
356 async_extent->start = start;
357 async_extent->ram_size = ram_size;
358 async_extent->compressed_size = compressed_size;
359 async_extent->pages = pages;
360 async_extent->nr_pages = nr_pages;
361 async_extent->compress_type = compress_type;
362 list_add_tail(&async_extent->list, &cow->extents);
363 return 0;
364 }
365
366 static inline int inode_need_compress(struct inode *inode)
367 {
368 struct btrfs_root *root = BTRFS_I(inode)->root;
369
370 /* force compress */
371 if (btrfs_test_opt(root, FORCE_COMPRESS))
372 return 1;
373 /* bad compression ratios */
374 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
375 return 0;
376 if (btrfs_test_opt(root, COMPRESS) ||
377 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
378 BTRFS_I(inode)->force_compress)
379 return 1;
380 return 0;
381 }
382
383 /*
384 * we create compressed extents in two phases. The first
385 * phase compresses a range of pages that have already been
386 * locked (both pages and state bits are locked).
387 *
388 * This is done inside an ordered work queue, and the compression
389 * is spread across many cpus. The actual IO submission is step
390 * two, and the ordered work queue takes care of making sure that
391 * happens in the same order things were put onto the queue by
392 * writepages and friends.
393 *
394 * If this code finds it can't get good compression, it puts an
395 * entry onto the work queue to write the uncompressed bytes. This
396 * makes sure that both compressed inodes and uncompressed inodes
397 * are written in the same order that the flusher thread sent them
398 * down.
399 */
400 static noinline void compress_file_range(struct inode *inode,
401 struct page *locked_page,
402 u64 start, u64 end,
403 struct async_cow *async_cow,
404 int *num_added)
405 {
406 struct btrfs_root *root = BTRFS_I(inode)->root;
407 u64 num_bytes;
408 u64 blocksize = root->sectorsize;
409 u64 actual_end;
410 u64 isize = i_size_read(inode);
411 int ret = 0;
412 struct page **pages = NULL;
413 unsigned long nr_pages;
414 unsigned long nr_pages_ret = 0;
415 unsigned long total_compressed = 0;
416 unsigned long total_in = 0;
417 unsigned long max_compressed = 128 * 1024;
418 unsigned long max_uncompressed = 128 * 1024;
419 int i;
420 int will_compress;
421 int compress_type = root->fs_info->compress_type;
422 int redirty = 0;
423
424 /* if this is a small write inside eof, kick off a defrag */
425 if ((end - start + 1) < 16 * 1024 &&
426 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
427 btrfs_add_inode_defrag(NULL, inode);
428
429 actual_end = min_t(u64, isize, end + 1);
430 again:
431 will_compress = 0;
432 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
433 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
434
435 /*
436 * we don't want to send crud past the end of i_size through
437 * compression, that's just a waste of CPU time. So, if the
438 * end of the file is before the start of our current
439 * requested range of bytes, we bail out to the uncompressed
440 * cleanup code that can deal with all of this.
441 *
442 * It isn't really the fastest way to fix things, but this is a
443 * very uncommon corner.
444 */
445 if (actual_end <= start)
446 goto cleanup_and_bail_uncompressed;
447
448 total_compressed = actual_end - start;
449
450 /*
451 * skip compression for a small file range(<=blocksize) that
452 * isn't an inline extent, since it dosen't save disk space at all.
453 */
454 if (total_compressed <= blocksize &&
455 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
456 goto cleanup_and_bail_uncompressed;
457
458 /* we want to make sure that amount of ram required to uncompress
459 * an extent is reasonable, so we limit the total size in ram
460 * of a compressed extent to 128k. This is a crucial number
461 * because it also controls how easily we can spread reads across
462 * cpus for decompression.
463 *
464 * We also want to make sure the amount of IO required to do
465 * a random read is reasonably small, so we limit the size of
466 * a compressed extent to 128k.
467 */
468 total_compressed = min(total_compressed, max_uncompressed);
469 num_bytes = ALIGN(end - start + 1, blocksize);
470 num_bytes = max(blocksize, num_bytes);
471 total_in = 0;
472 ret = 0;
473
474 /*
475 * we do compression for mount -o compress and when the
476 * inode has not been flagged as nocompress. This flag can
477 * change at any time if we discover bad compression ratios.
478 */
479 if (inode_need_compress(inode)) {
480 WARN_ON(pages);
481 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
482 if (!pages) {
483 /* just bail out to the uncompressed code */
484 goto cont;
485 }
486
487 if (BTRFS_I(inode)->force_compress)
488 compress_type = BTRFS_I(inode)->force_compress;
489
490 /*
491 * we need to call clear_page_dirty_for_io on each
492 * page in the range. Otherwise applications with the file
493 * mmap'd can wander in and change the page contents while
494 * we are compressing them.
495 *
496 * If the compression fails for any reason, we set the pages
497 * dirty again later on.
498 */
499 extent_range_clear_dirty_for_io(inode, start, end);
500 redirty = 1;
501 ret = btrfs_compress_pages(compress_type,
502 inode->i_mapping, start,
503 total_compressed, pages,
504 nr_pages, &nr_pages_ret,
505 &total_in,
506 &total_compressed,
507 max_compressed);
508
509 if (!ret) {
510 unsigned long offset = total_compressed &
511 (PAGE_CACHE_SIZE - 1);
512 struct page *page = pages[nr_pages_ret - 1];
513 char *kaddr;
514
515 /* zero the tail end of the last page, we might be
516 * sending it down to disk
517 */
518 if (offset) {
519 kaddr = kmap_atomic(page);
520 memset(kaddr + offset, 0,
521 PAGE_CACHE_SIZE - offset);
522 kunmap_atomic(kaddr);
523 }
524 will_compress = 1;
525 }
526 }
527 cont:
528 if (start == 0) {
529 /* lets try to make an inline extent */
530 if (ret || total_in < (actual_end - start)) {
531 /* we didn't compress the entire range, try
532 * to make an uncompressed inline extent.
533 */
534 ret = cow_file_range_inline(root, inode, start, end,
535 0, 0, NULL);
536 } else {
537 /* try making a compressed inline extent */
538 ret = cow_file_range_inline(root, inode, start, end,
539 total_compressed,
540 compress_type, pages);
541 }
542 if (ret <= 0) {
543 unsigned long clear_flags = EXTENT_DELALLOC |
544 EXTENT_DEFRAG;
545 unsigned long page_error_op;
546
547 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
548 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
549
550 /*
551 * inline extent creation worked or returned error,
552 * we don't need to create any more async work items.
553 * Unlock and free up our temp pages.
554 */
555 extent_clear_unlock_delalloc(inode, start, end, NULL,
556 clear_flags, PAGE_UNLOCK |
557 PAGE_CLEAR_DIRTY |
558 PAGE_SET_WRITEBACK |
559 page_error_op |
560 PAGE_END_WRITEBACK);
561 goto free_pages_out;
562 }
563 }
564
565 if (will_compress) {
566 /*
567 * we aren't doing an inline extent round the compressed size
568 * up to a block size boundary so the allocator does sane
569 * things
570 */
571 total_compressed = ALIGN(total_compressed, blocksize);
572
573 /*
574 * one last check to make sure the compression is really a
575 * win, compare the page count read with the blocks on disk
576 */
577 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
578 if (total_compressed >= total_in) {
579 will_compress = 0;
580 } else {
581 num_bytes = total_in;
582 }
583 }
584 if (!will_compress && pages) {
585 /*
586 * the compression code ran but failed to make things smaller,
587 * free any pages it allocated and our page pointer array
588 */
589 for (i = 0; i < nr_pages_ret; i++) {
590 WARN_ON(pages[i]->mapping);
591 page_cache_release(pages[i]);
592 }
593 kfree(pages);
594 pages = NULL;
595 total_compressed = 0;
596 nr_pages_ret = 0;
597
598 /* flag the file so we don't compress in the future */
599 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
600 !(BTRFS_I(inode)->force_compress)) {
601 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
602 }
603 }
604 if (will_compress) {
605 *num_added += 1;
606
607 /* the async work queues will take care of doing actual
608 * allocation on disk for these compressed pages,
609 * and will submit them to the elevator.
610 */
611 add_async_extent(async_cow, start, num_bytes,
612 total_compressed, pages, nr_pages_ret,
613 compress_type);
614
615 if (start + num_bytes < end) {
616 start += num_bytes;
617 pages = NULL;
618 cond_resched();
619 goto again;
620 }
621 } else {
622 cleanup_and_bail_uncompressed:
623 /*
624 * No compression, but we still need to write the pages in
625 * the file we've been given so far. redirty the locked
626 * page if it corresponds to our extent and set things up
627 * for the async work queue to run cow_file_range to do
628 * the normal delalloc dance
629 */
630 if (page_offset(locked_page) >= start &&
631 page_offset(locked_page) <= end) {
632 __set_page_dirty_nobuffers(locked_page);
633 /* unlocked later on in the async handlers */
634 }
635 if (redirty)
636 extent_range_redirty_for_io(inode, start, end);
637 add_async_extent(async_cow, start, end - start + 1,
638 0, NULL, 0, BTRFS_COMPRESS_NONE);
639 *num_added += 1;
640 }
641
642 return;
643
644 free_pages_out:
645 for (i = 0; i < nr_pages_ret; i++) {
646 WARN_ON(pages[i]->mapping);
647 page_cache_release(pages[i]);
648 }
649 kfree(pages);
650 }
651
652 static void free_async_extent_pages(struct async_extent *async_extent)
653 {
654 int i;
655
656 if (!async_extent->pages)
657 return;
658
659 for (i = 0; i < async_extent->nr_pages; i++) {
660 WARN_ON(async_extent->pages[i]->mapping);
661 page_cache_release(async_extent->pages[i]);
662 }
663 kfree(async_extent->pages);
664 async_extent->nr_pages = 0;
665 async_extent->pages = NULL;
666 }
667
668 /*
669 * phase two of compressed writeback. This is the ordered portion
670 * of the code, which only gets called in the order the work was
671 * queued. We walk all the async extents created by compress_file_range
672 * and send them down to the disk.
673 */
674 static noinline void submit_compressed_extents(struct inode *inode,
675 struct async_cow *async_cow)
676 {
677 struct async_extent *async_extent;
678 u64 alloc_hint = 0;
679 struct btrfs_key ins;
680 struct extent_map *em;
681 struct btrfs_root *root = BTRFS_I(inode)->root;
682 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
683 struct extent_io_tree *io_tree;
684 int ret = 0;
685
686 again:
687 while (!list_empty(&async_cow->extents)) {
688 async_extent = list_entry(async_cow->extents.next,
689 struct async_extent, list);
690 list_del(&async_extent->list);
691
692 io_tree = &BTRFS_I(inode)->io_tree;
693
694 retry:
695 /* did the compression code fall back to uncompressed IO? */
696 if (!async_extent->pages) {
697 int page_started = 0;
698 unsigned long nr_written = 0;
699
700 lock_extent(io_tree, async_extent->start,
701 async_extent->start +
702 async_extent->ram_size - 1);
703
704 /* allocate blocks */
705 ret = cow_file_range(inode, async_cow->locked_page,
706 async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1,
709 &page_started, &nr_written, 0);
710
711 /* JDM XXX */
712
713 /*
714 * if page_started, cow_file_range inserted an
715 * inline extent and took care of all the unlocking
716 * and IO for us. Otherwise, we need to submit
717 * all those pages down to the drive.
718 */
719 if (!page_started && !ret)
720 extent_write_locked_range(io_tree,
721 inode, async_extent->start,
722 async_extent->start +
723 async_extent->ram_size - 1,
724 btrfs_get_extent,
725 WB_SYNC_ALL);
726 else if (ret)
727 unlock_page(async_cow->locked_page);
728 kfree(async_extent);
729 cond_resched();
730 continue;
731 }
732
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start + async_extent->ram_size - 1);
735
736 ret = btrfs_reserve_extent(root,
737 async_extent->compressed_size,
738 async_extent->compressed_size,
739 0, alloc_hint, &ins, 1, 1);
740 if (ret) {
741 free_async_extent_pages(async_extent);
742
743 if (ret == -ENOSPC) {
744 unlock_extent(io_tree, async_extent->start,
745 async_extent->start +
746 async_extent->ram_size - 1);
747
748 /*
749 * we need to redirty the pages if we decide to
750 * fallback to uncompressed IO, otherwise we
751 * will not submit these pages down to lower
752 * layers.
753 */
754 extent_range_redirty_for_io(inode,
755 async_extent->start,
756 async_extent->start +
757 async_extent->ram_size - 1);
758
759 goto retry;
760 }
761 goto out_free;
762 }
763 /*
764 * here we're doing allocation and writeback of the
765 * compressed pages
766 */
767 btrfs_drop_extent_cache(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1, 0);
770
771 em = alloc_extent_map();
772 if (!em) {
773 ret = -ENOMEM;
774 goto out_free_reserve;
775 }
776 em->start = async_extent->start;
777 em->len = async_extent->ram_size;
778 em->orig_start = em->start;
779 em->mod_start = em->start;
780 em->mod_len = em->len;
781
782 em->block_start = ins.objectid;
783 em->block_len = ins.offset;
784 em->orig_block_len = ins.offset;
785 em->ram_bytes = async_extent->ram_size;
786 em->bdev = root->fs_info->fs_devices->latest_bdev;
787 em->compress_type = async_extent->compress_type;
788 set_bit(EXTENT_FLAG_PINNED, &em->flags);
789 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
790 em->generation = -1;
791
792 while (1) {
793 write_lock(&em_tree->lock);
794 ret = add_extent_mapping(em_tree, em, 1);
795 write_unlock(&em_tree->lock);
796 if (ret != -EEXIST) {
797 free_extent_map(em);
798 break;
799 }
800 btrfs_drop_extent_cache(inode, async_extent->start,
801 async_extent->start +
802 async_extent->ram_size - 1, 0);
803 }
804
805 if (ret)
806 goto out_free_reserve;
807
808 ret = btrfs_add_ordered_extent_compress(inode,
809 async_extent->start,
810 ins.objectid,
811 async_extent->ram_size,
812 ins.offset,
813 BTRFS_ORDERED_COMPRESSED,
814 async_extent->compress_type);
815 if (ret) {
816 btrfs_drop_extent_cache(inode, async_extent->start,
817 async_extent->start +
818 async_extent->ram_size - 1, 0);
819 goto out_free_reserve;
820 }
821
822 /*
823 * clear dirty, set writeback and unlock the pages.
824 */
825 extent_clear_unlock_delalloc(inode, async_extent->start,
826 async_extent->start +
827 async_extent->ram_size - 1,
828 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
829 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
830 PAGE_SET_WRITEBACK);
831 ret = btrfs_submit_compressed_write(inode,
832 async_extent->start,
833 async_extent->ram_size,
834 ins.objectid,
835 ins.offset, async_extent->pages,
836 async_extent->nr_pages);
837 if (ret) {
838 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
839 struct page *p = async_extent->pages[0];
840 const u64 start = async_extent->start;
841 const u64 end = start + async_extent->ram_size - 1;
842
843 p->mapping = inode->i_mapping;
844 tree->ops->writepage_end_io_hook(p, start, end,
845 NULL, 0);
846 p->mapping = NULL;
847 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
848 PAGE_END_WRITEBACK |
849 PAGE_SET_ERROR);
850 free_async_extent_pages(async_extent);
851 }
852 alloc_hint = ins.objectid + ins.offset;
853 kfree(async_extent);
854 cond_resched();
855 }
856 return;
857 out_free_reserve:
858 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
859 out_free:
860 extent_clear_unlock_delalloc(inode, async_extent->start,
861 async_extent->start +
862 async_extent->ram_size - 1,
863 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
864 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
865 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
866 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
867 PAGE_SET_ERROR);
868 free_async_extent_pages(async_extent);
869 kfree(async_extent);
870 goto again;
871 }
872
873 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
874 u64 num_bytes)
875 {
876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
877 struct extent_map *em;
878 u64 alloc_hint = 0;
879
880 read_lock(&em_tree->lock);
881 em = search_extent_mapping(em_tree, start, num_bytes);
882 if (em) {
883 /*
884 * if block start isn't an actual block number then find the
885 * first block in this inode and use that as a hint. If that
886 * block is also bogus then just don't worry about it.
887 */
888 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
889 free_extent_map(em);
890 em = search_extent_mapping(em_tree, 0, 0);
891 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
892 alloc_hint = em->block_start;
893 if (em)
894 free_extent_map(em);
895 } else {
896 alloc_hint = em->block_start;
897 free_extent_map(em);
898 }
899 }
900 read_unlock(&em_tree->lock);
901
902 return alloc_hint;
903 }
904
905 /*
906 * when extent_io.c finds a delayed allocation range in the file,
907 * the call backs end up in this code. The basic idea is to
908 * allocate extents on disk for the range, and create ordered data structs
909 * in ram to track those extents.
910 *
911 * locked_page is the page that writepage had locked already. We use
912 * it to make sure we don't do extra locks or unlocks.
913 *
914 * *page_started is set to one if we unlock locked_page and do everything
915 * required to start IO on it. It may be clean and already done with
916 * IO when we return.
917 */
918 static noinline int cow_file_range(struct inode *inode,
919 struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written,
922 int unlock)
923 {
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 u64 alloc_hint = 0;
926 u64 num_bytes;
927 unsigned long ram_size;
928 u64 disk_num_bytes;
929 u64 cur_alloc_size;
930 u64 blocksize = root->sectorsize;
931 struct btrfs_key ins;
932 struct extent_map *em;
933 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
934 int ret = 0;
935
936 if (btrfs_is_free_space_inode(inode)) {
937 WARN_ON_ONCE(1);
938 ret = -EINVAL;
939 goto out_unlock;
940 }
941
942 num_bytes = ALIGN(end - start + 1, blocksize);
943 num_bytes = max(blocksize, num_bytes);
944 disk_num_bytes = num_bytes;
945
946 /* if this is a small write inside eof, kick off defrag */
947 if (num_bytes < 64 * 1024 &&
948 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
949 btrfs_add_inode_defrag(NULL, inode);
950
951 if (start == 0) {
952 /* lets try to make an inline extent */
953 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
954 NULL);
955 if (ret == 0) {
956 extent_clear_unlock_delalloc(inode, start, end, NULL,
957 EXTENT_LOCKED | EXTENT_DELALLOC |
958 EXTENT_DEFRAG, PAGE_UNLOCK |
959 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
960 PAGE_END_WRITEBACK);
961
962 *nr_written = *nr_written +
963 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
964 *page_started = 1;
965 goto out;
966 } else if (ret < 0) {
967 goto out_unlock;
968 }
969 }
970
971 BUG_ON(disk_num_bytes >
972 btrfs_super_total_bytes(root->fs_info->super_copy));
973
974 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
975 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
976
977 while (disk_num_bytes > 0) {
978 unsigned long op;
979
980 cur_alloc_size = disk_num_bytes;
981 ret = btrfs_reserve_extent(root, cur_alloc_size,
982 root->sectorsize, 0, alloc_hint,
983 &ins, 1, 1);
984 if (ret < 0)
985 goto out_unlock;
986
987 em = alloc_extent_map();
988 if (!em) {
989 ret = -ENOMEM;
990 goto out_reserve;
991 }
992 em->start = start;
993 em->orig_start = em->start;
994 ram_size = ins.offset;
995 em->len = ins.offset;
996 em->mod_start = em->start;
997 em->mod_len = em->len;
998
999 em->block_start = ins.objectid;
1000 em->block_len = ins.offset;
1001 em->orig_block_len = ins.offset;
1002 em->ram_bytes = ram_size;
1003 em->bdev = root->fs_info->fs_devices->latest_bdev;
1004 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1005 em->generation = -1;
1006
1007 while (1) {
1008 write_lock(&em_tree->lock);
1009 ret = add_extent_mapping(em_tree, em, 1);
1010 write_unlock(&em_tree->lock);
1011 if (ret != -EEXIST) {
1012 free_extent_map(em);
1013 break;
1014 }
1015 btrfs_drop_extent_cache(inode, start,
1016 start + ram_size - 1, 0);
1017 }
1018 if (ret)
1019 goto out_reserve;
1020
1021 cur_alloc_size = ins.offset;
1022 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1023 ram_size, cur_alloc_size, 0);
1024 if (ret)
1025 goto out_drop_extent_cache;
1026
1027 if (root->root_key.objectid ==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1029 ret = btrfs_reloc_clone_csums(inode, start,
1030 cur_alloc_size);
1031 if (ret)
1032 goto out_drop_extent_cache;
1033 }
1034
1035 if (disk_num_bytes < cur_alloc_size)
1036 break;
1037
1038 /* we're not doing compressed IO, don't unlock the first
1039 * page (which the caller expects to stay locked), don't
1040 * clear any dirty bits and don't set any writeback bits
1041 *
1042 * Do set the Private2 bit so we know this page was properly
1043 * setup for writepage
1044 */
1045 op = unlock ? PAGE_UNLOCK : 0;
1046 op |= PAGE_SET_PRIVATE2;
1047
1048 extent_clear_unlock_delalloc(inode, start,
1049 start + ram_size - 1, locked_page,
1050 EXTENT_LOCKED | EXTENT_DELALLOC,
1051 op);
1052 disk_num_bytes -= cur_alloc_size;
1053 num_bytes -= cur_alloc_size;
1054 alloc_hint = ins.objectid + ins.offset;
1055 start += cur_alloc_size;
1056 }
1057 out:
1058 return ret;
1059
1060 out_drop_extent_cache:
1061 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1062 out_reserve:
1063 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1064 out_unlock:
1065 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1066 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1067 EXTENT_DELALLOC | EXTENT_DEFRAG,
1068 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1069 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1070 goto out;
1071 }
1072
1073 /*
1074 * work queue call back to started compression on a file and pages
1075 */
1076 static noinline void async_cow_start(struct btrfs_work *work)
1077 {
1078 struct async_cow *async_cow;
1079 int num_added = 0;
1080 async_cow = container_of(work, struct async_cow, work);
1081
1082 compress_file_range(async_cow->inode, async_cow->locked_page,
1083 async_cow->start, async_cow->end, async_cow,
1084 &num_added);
1085 if (num_added == 0) {
1086 btrfs_add_delayed_iput(async_cow->inode);
1087 async_cow->inode = NULL;
1088 }
1089 }
1090
1091 /*
1092 * work queue call back to submit previously compressed pages
1093 */
1094 static noinline void async_cow_submit(struct btrfs_work *work)
1095 {
1096 struct async_cow *async_cow;
1097 struct btrfs_root *root;
1098 unsigned long nr_pages;
1099
1100 async_cow = container_of(work, struct async_cow, work);
1101
1102 root = async_cow->root;
1103 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1104 PAGE_CACHE_SHIFT;
1105
1106 /*
1107 * atomic_sub_return implies a barrier for waitqueue_active
1108 */
1109 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1110 5 * 1024 * 1024 &&
1111 waitqueue_active(&root->fs_info->async_submit_wait))
1112 wake_up(&root->fs_info->async_submit_wait);
1113
1114 if (async_cow->inode)
1115 submit_compressed_extents(async_cow->inode, async_cow);
1116 }
1117
1118 static noinline void async_cow_free(struct btrfs_work *work)
1119 {
1120 struct async_cow *async_cow;
1121 async_cow = container_of(work, struct async_cow, work);
1122 if (async_cow->inode)
1123 btrfs_add_delayed_iput(async_cow->inode);
1124 kfree(async_cow);
1125 }
1126
1127 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1128 u64 start, u64 end, int *page_started,
1129 unsigned long *nr_written)
1130 {
1131 struct async_cow *async_cow;
1132 struct btrfs_root *root = BTRFS_I(inode)->root;
1133 unsigned long nr_pages;
1134 u64 cur_end;
1135 int limit = 10 * 1024 * 1024;
1136
1137 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1138 1, 0, NULL, GFP_NOFS);
1139 while (start < end) {
1140 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1141 BUG_ON(!async_cow); /* -ENOMEM */
1142 async_cow->inode = igrab(inode);
1143 async_cow->root = root;
1144 async_cow->locked_page = locked_page;
1145 async_cow->start = start;
1146
1147 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1148 !btrfs_test_opt(root, FORCE_COMPRESS))
1149 cur_end = end;
1150 else
1151 cur_end = min(end, start + 512 * 1024 - 1);
1152
1153 async_cow->end = cur_end;
1154 INIT_LIST_HEAD(&async_cow->extents);
1155
1156 btrfs_init_work(&async_cow->work,
1157 btrfs_delalloc_helper,
1158 async_cow_start, async_cow_submit,
1159 async_cow_free);
1160
1161 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1162 PAGE_CACHE_SHIFT;
1163 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1164
1165 btrfs_queue_work(root->fs_info->delalloc_workers,
1166 &async_cow->work);
1167
1168 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1169 wait_event(root->fs_info->async_submit_wait,
1170 (atomic_read(&root->fs_info->async_delalloc_pages) <
1171 limit));
1172 }
1173
1174 while (atomic_read(&root->fs_info->async_submit_draining) &&
1175 atomic_read(&root->fs_info->async_delalloc_pages)) {
1176 wait_event(root->fs_info->async_submit_wait,
1177 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1178 0));
1179 }
1180
1181 *nr_written += nr_pages;
1182 start = cur_end + 1;
1183 }
1184 *page_started = 1;
1185 return 0;
1186 }
1187
1188 static noinline int csum_exist_in_range(struct btrfs_root *root,
1189 u64 bytenr, u64 num_bytes)
1190 {
1191 int ret;
1192 struct btrfs_ordered_sum *sums;
1193 LIST_HEAD(list);
1194
1195 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1196 bytenr + num_bytes - 1, &list, 0);
1197 if (ret == 0 && list_empty(&list))
1198 return 0;
1199
1200 while (!list_empty(&list)) {
1201 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1202 list_del(&sums->list);
1203 kfree(sums);
1204 }
1205 return 1;
1206 }
1207
1208 /*
1209 * when nowcow writeback call back. This checks for snapshots or COW copies
1210 * of the extents that exist in the file, and COWs the file as required.
1211 *
1212 * If no cow copies or snapshots exist, we write directly to the existing
1213 * blocks on disk
1214 */
1215 static noinline int run_delalloc_nocow(struct inode *inode,
1216 struct page *locked_page,
1217 u64 start, u64 end, int *page_started, int force,
1218 unsigned long *nr_written)
1219 {
1220 struct btrfs_root *root = BTRFS_I(inode)->root;
1221 struct btrfs_trans_handle *trans;
1222 struct extent_buffer *leaf;
1223 struct btrfs_path *path;
1224 struct btrfs_file_extent_item *fi;
1225 struct btrfs_key found_key;
1226 u64 cow_start;
1227 u64 cur_offset;
1228 u64 extent_end;
1229 u64 extent_offset;
1230 u64 disk_bytenr;
1231 u64 num_bytes;
1232 u64 disk_num_bytes;
1233 u64 ram_bytes;
1234 int extent_type;
1235 int ret, err;
1236 int type;
1237 int nocow;
1238 int check_prev = 1;
1239 bool nolock;
1240 u64 ino = btrfs_ino(inode);
1241
1242 path = btrfs_alloc_path();
1243 if (!path) {
1244 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1245 EXTENT_LOCKED | EXTENT_DELALLOC |
1246 EXTENT_DO_ACCOUNTING |
1247 EXTENT_DEFRAG, PAGE_UNLOCK |
1248 PAGE_CLEAR_DIRTY |
1249 PAGE_SET_WRITEBACK |
1250 PAGE_END_WRITEBACK);
1251 return -ENOMEM;
1252 }
1253
1254 nolock = btrfs_is_free_space_inode(inode);
1255
1256 if (nolock)
1257 trans = btrfs_join_transaction_nolock(root);
1258 else
1259 trans = btrfs_join_transaction(root);
1260
1261 if (IS_ERR(trans)) {
1262 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1263 EXTENT_LOCKED | EXTENT_DELALLOC |
1264 EXTENT_DO_ACCOUNTING |
1265 EXTENT_DEFRAG, PAGE_UNLOCK |
1266 PAGE_CLEAR_DIRTY |
1267 PAGE_SET_WRITEBACK |
1268 PAGE_END_WRITEBACK);
1269 btrfs_free_path(path);
1270 return PTR_ERR(trans);
1271 }
1272
1273 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1274
1275 cow_start = (u64)-1;
1276 cur_offset = start;
1277 while (1) {
1278 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1279 cur_offset, 0);
1280 if (ret < 0)
1281 goto error;
1282 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &found_key,
1285 path->slots[0] - 1);
1286 if (found_key.objectid == ino &&
1287 found_key.type == BTRFS_EXTENT_DATA_KEY)
1288 path->slots[0]--;
1289 }
1290 check_prev = 0;
1291 next_slot:
1292 leaf = path->nodes[0];
1293 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1294 ret = btrfs_next_leaf(root, path);
1295 if (ret < 0)
1296 goto error;
1297 if (ret > 0)
1298 break;
1299 leaf = path->nodes[0];
1300 }
1301
1302 nocow = 0;
1303 disk_bytenr = 0;
1304 num_bytes = 0;
1305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1306
1307 if (found_key.objectid > ino)
1308 break;
1309 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1310 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1311 path->slots[0]++;
1312 goto next_slot;
1313 }
1314 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1315 found_key.offset > end)
1316 break;
1317
1318 if (found_key.offset > cur_offset) {
1319 extent_end = found_key.offset;
1320 extent_type = 0;
1321 goto out_check;
1322 }
1323
1324 fi = btrfs_item_ptr(leaf, path->slots[0],
1325 struct btrfs_file_extent_item);
1326 extent_type = btrfs_file_extent_type(leaf, fi);
1327
1328 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1329 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1330 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1331 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1332 extent_offset = btrfs_file_extent_offset(leaf, fi);
1333 extent_end = found_key.offset +
1334 btrfs_file_extent_num_bytes(leaf, fi);
1335 disk_num_bytes =
1336 btrfs_file_extent_disk_num_bytes(leaf, fi);
1337 if (extent_end <= start) {
1338 path->slots[0]++;
1339 goto next_slot;
1340 }
1341 if (disk_bytenr == 0)
1342 goto out_check;
1343 if (btrfs_file_extent_compression(leaf, fi) ||
1344 btrfs_file_extent_encryption(leaf, fi) ||
1345 btrfs_file_extent_other_encoding(leaf, fi))
1346 goto out_check;
1347 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1348 goto out_check;
1349 if (btrfs_extent_readonly(root, disk_bytenr))
1350 goto out_check;
1351 if (btrfs_cross_ref_exist(trans, root, ino,
1352 found_key.offset -
1353 extent_offset, disk_bytenr))
1354 goto out_check;
1355 disk_bytenr += extent_offset;
1356 disk_bytenr += cur_offset - found_key.offset;
1357 num_bytes = min(end + 1, extent_end) - cur_offset;
1358 /*
1359 * if there are pending snapshots for this root,
1360 * we fall into common COW way.
1361 */
1362 if (!nolock) {
1363 err = btrfs_start_write_no_snapshoting(root);
1364 if (!err)
1365 goto out_check;
1366 }
1367 /*
1368 * force cow if csum exists in the range.
1369 * this ensure that csum for a given extent are
1370 * either valid or do not exist.
1371 */
1372 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1373 goto out_check;
1374 nocow = 1;
1375 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1376 extent_end = found_key.offset +
1377 btrfs_file_extent_inline_len(leaf,
1378 path->slots[0], fi);
1379 extent_end = ALIGN(extent_end, root->sectorsize);
1380 } else {
1381 BUG_ON(1);
1382 }
1383 out_check:
1384 if (extent_end <= start) {
1385 path->slots[0]++;
1386 if (!nolock && nocow)
1387 btrfs_end_write_no_snapshoting(root);
1388 goto next_slot;
1389 }
1390 if (!nocow) {
1391 if (cow_start == (u64)-1)
1392 cow_start = cur_offset;
1393 cur_offset = extent_end;
1394 if (cur_offset > end)
1395 break;
1396 path->slots[0]++;
1397 goto next_slot;
1398 }
1399
1400 btrfs_release_path(path);
1401 if (cow_start != (u64)-1) {
1402 ret = cow_file_range(inode, locked_page,
1403 cow_start, found_key.offset - 1,
1404 page_started, nr_written, 1);
1405 if (ret) {
1406 if (!nolock && nocow)
1407 btrfs_end_write_no_snapshoting(root);
1408 goto error;
1409 }
1410 cow_start = (u64)-1;
1411 }
1412
1413 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1414 struct extent_map *em;
1415 struct extent_map_tree *em_tree;
1416 em_tree = &BTRFS_I(inode)->extent_tree;
1417 em = alloc_extent_map();
1418 BUG_ON(!em); /* -ENOMEM */
1419 em->start = cur_offset;
1420 em->orig_start = found_key.offset - extent_offset;
1421 em->len = num_bytes;
1422 em->block_len = num_bytes;
1423 em->block_start = disk_bytenr;
1424 em->orig_block_len = disk_num_bytes;
1425 em->ram_bytes = ram_bytes;
1426 em->bdev = root->fs_info->fs_devices->latest_bdev;
1427 em->mod_start = em->start;
1428 em->mod_len = em->len;
1429 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1430 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1431 em->generation = -1;
1432 while (1) {
1433 write_lock(&em_tree->lock);
1434 ret = add_extent_mapping(em_tree, em, 1);
1435 write_unlock(&em_tree->lock);
1436 if (ret != -EEXIST) {
1437 free_extent_map(em);
1438 break;
1439 }
1440 btrfs_drop_extent_cache(inode, em->start,
1441 em->start + em->len - 1, 0);
1442 }
1443 type = BTRFS_ORDERED_PREALLOC;
1444 } else {
1445 type = BTRFS_ORDERED_NOCOW;
1446 }
1447
1448 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1449 num_bytes, num_bytes, type);
1450 BUG_ON(ret); /* -ENOMEM */
1451
1452 if (root->root_key.objectid ==
1453 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1454 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1455 num_bytes);
1456 if (ret) {
1457 if (!nolock && nocow)
1458 btrfs_end_write_no_snapshoting(root);
1459 goto error;
1460 }
1461 }
1462
1463 extent_clear_unlock_delalloc(inode, cur_offset,
1464 cur_offset + num_bytes - 1,
1465 locked_page, EXTENT_LOCKED |
1466 EXTENT_DELALLOC, PAGE_UNLOCK |
1467 PAGE_SET_PRIVATE2);
1468 if (!nolock && nocow)
1469 btrfs_end_write_no_snapshoting(root);
1470 cur_offset = extent_end;
1471 if (cur_offset > end)
1472 break;
1473 }
1474 btrfs_release_path(path);
1475
1476 if (cur_offset <= end && cow_start == (u64)-1) {
1477 cow_start = cur_offset;
1478 cur_offset = end;
1479 }
1480
1481 if (cow_start != (u64)-1) {
1482 ret = cow_file_range(inode, locked_page, cow_start, end,
1483 page_started, nr_written, 1);
1484 if (ret)
1485 goto error;
1486 }
1487
1488 error:
1489 err = btrfs_end_transaction(trans, root);
1490 if (!ret)
1491 ret = err;
1492
1493 if (ret && cur_offset < end)
1494 extent_clear_unlock_delalloc(inode, cur_offset, end,
1495 locked_page, EXTENT_LOCKED |
1496 EXTENT_DELALLOC | EXTENT_DEFRAG |
1497 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1498 PAGE_CLEAR_DIRTY |
1499 PAGE_SET_WRITEBACK |
1500 PAGE_END_WRITEBACK);
1501 btrfs_free_path(path);
1502 return ret;
1503 }
1504
1505 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1506 {
1507
1508 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1509 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1510 return 0;
1511
1512 /*
1513 * @defrag_bytes is a hint value, no spinlock held here,
1514 * if is not zero, it means the file is defragging.
1515 * Force cow if given extent needs to be defragged.
1516 */
1517 if (BTRFS_I(inode)->defrag_bytes &&
1518 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1519 EXTENT_DEFRAG, 0, NULL))
1520 return 1;
1521
1522 return 0;
1523 }
1524
1525 /*
1526 * extent_io.c call back to do delayed allocation processing
1527 */
1528 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1529 u64 start, u64 end, int *page_started,
1530 unsigned long *nr_written)
1531 {
1532 int ret;
1533 int force_cow = need_force_cow(inode, start, end);
1534
1535 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1536 ret = run_delalloc_nocow(inode, locked_page, start, end,
1537 page_started, 1, nr_written);
1538 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1539 ret = run_delalloc_nocow(inode, locked_page, start, end,
1540 page_started, 0, nr_written);
1541 } else if (!inode_need_compress(inode)) {
1542 ret = cow_file_range(inode, locked_page, start, end,
1543 page_started, nr_written, 1);
1544 } else {
1545 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1546 &BTRFS_I(inode)->runtime_flags);
1547 ret = cow_file_range_async(inode, locked_page, start, end,
1548 page_started, nr_written);
1549 }
1550 return ret;
1551 }
1552
1553 static void btrfs_split_extent_hook(struct inode *inode,
1554 struct extent_state *orig, u64 split)
1555 {
1556 u64 size;
1557
1558 /* not delalloc, ignore it */
1559 if (!(orig->state & EXTENT_DELALLOC))
1560 return;
1561
1562 size = orig->end - orig->start + 1;
1563 if (size > BTRFS_MAX_EXTENT_SIZE) {
1564 u64 num_extents;
1565 u64 new_size;
1566
1567 /*
1568 * See the explanation in btrfs_merge_extent_hook, the same
1569 * applies here, just in reverse.
1570 */
1571 new_size = orig->end - split + 1;
1572 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1573 BTRFS_MAX_EXTENT_SIZE);
1574 new_size = split - orig->start;
1575 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1576 BTRFS_MAX_EXTENT_SIZE);
1577 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1578 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1579 return;
1580 }
1581
1582 spin_lock(&BTRFS_I(inode)->lock);
1583 BTRFS_I(inode)->outstanding_extents++;
1584 spin_unlock(&BTRFS_I(inode)->lock);
1585 }
1586
1587 /*
1588 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1589 * extents so we can keep track of new extents that are just merged onto old
1590 * extents, such as when we are doing sequential writes, so we can properly
1591 * account for the metadata space we'll need.
1592 */
1593 static void btrfs_merge_extent_hook(struct inode *inode,
1594 struct extent_state *new,
1595 struct extent_state *other)
1596 {
1597 u64 new_size, old_size;
1598 u64 num_extents;
1599
1600 /* not delalloc, ignore it */
1601 if (!(other->state & EXTENT_DELALLOC))
1602 return;
1603
1604 if (new->start > other->start)
1605 new_size = new->end - other->start + 1;
1606 else
1607 new_size = other->end - new->start + 1;
1608
1609 /* we're not bigger than the max, unreserve the space and go */
1610 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1611 spin_lock(&BTRFS_I(inode)->lock);
1612 BTRFS_I(inode)->outstanding_extents--;
1613 spin_unlock(&BTRFS_I(inode)->lock);
1614 return;
1615 }
1616
1617 /*
1618 * We have to add up either side to figure out how many extents were
1619 * accounted for before we merged into one big extent. If the number of
1620 * extents we accounted for is <= the amount we need for the new range
1621 * then we can return, otherwise drop. Think of it like this
1622 *
1623 * [ 4k][MAX_SIZE]
1624 *
1625 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1626 * need 2 outstanding extents, on one side we have 1 and the other side
1627 * we have 1 so they are == and we can return. But in this case
1628 *
1629 * [MAX_SIZE+4k][MAX_SIZE+4k]
1630 *
1631 * Each range on their own accounts for 2 extents, but merged together
1632 * they are only 3 extents worth of accounting, so we need to drop in
1633 * this case.
1634 */
1635 old_size = other->end - other->start + 1;
1636 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1637 BTRFS_MAX_EXTENT_SIZE);
1638 old_size = new->end - new->start + 1;
1639 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1640 BTRFS_MAX_EXTENT_SIZE);
1641
1642 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1644 return;
1645
1646 spin_lock(&BTRFS_I(inode)->lock);
1647 BTRFS_I(inode)->outstanding_extents--;
1648 spin_unlock(&BTRFS_I(inode)->lock);
1649 }
1650
1651 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1652 struct inode *inode)
1653 {
1654 spin_lock(&root->delalloc_lock);
1655 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1656 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1657 &root->delalloc_inodes);
1658 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1659 &BTRFS_I(inode)->runtime_flags);
1660 root->nr_delalloc_inodes++;
1661 if (root->nr_delalloc_inodes == 1) {
1662 spin_lock(&root->fs_info->delalloc_root_lock);
1663 BUG_ON(!list_empty(&root->delalloc_root));
1664 list_add_tail(&root->delalloc_root,
1665 &root->fs_info->delalloc_roots);
1666 spin_unlock(&root->fs_info->delalloc_root_lock);
1667 }
1668 }
1669 spin_unlock(&root->delalloc_lock);
1670 }
1671
1672 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1673 struct inode *inode)
1674 {
1675 spin_lock(&root->delalloc_lock);
1676 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1677 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1678 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1679 &BTRFS_I(inode)->runtime_flags);
1680 root->nr_delalloc_inodes--;
1681 if (!root->nr_delalloc_inodes) {
1682 spin_lock(&root->fs_info->delalloc_root_lock);
1683 BUG_ON(list_empty(&root->delalloc_root));
1684 list_del_init(&root->delalloc_root);
1685 spin_unlock(&root->fs_info->delalloc_root_lock);
1686 }
1687 }
1688 spin_unlock(&root->delalloc_lock);
1689 }
1690
1691 /*
1692 * extent_io.c set_bit_hook, used to track delayed allocation
1693 * bytes in this file, and to maintain the list of inodes that
1694 * have pending delalloc work to be done.
1695 */
1696 static void btrfs_set_bit_hook(struct inode *inode,
1697 struct extent_state *state, unsigned *bits)
1698 {
1699
1700 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1701 WARN_ON(1);
1702 /*
1703 * set_bit and clear bit hooks normally require _irqsave/restore
1704 * but in this case, we are only testing for the DELALLOC
1705 * bit, which is only set or cleared with irqs on
1706 */
1707 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1708 struct btrfs_root *root = BTRFS_I(inode)->root;
1709 u64 len = state->end + 1 - state->start;
1710 bool do_list = !btrfs_is_free_space_inode(inode);
1711
1712 if (*bits & EXTENT_FIRST_DELALLOC) {
1713 *bits &= ~EXTENT_FIRST_DELALLOC;
1714 } else {
1715 spin_lock(&BTRFS_I(inode)->lock);
1716 BTRFS_I(inode)->outstanding_extents++;
1717 spin_unlock(&BTRFS_I(inode)->lock);
1718 }
1719
1720 /* For sanity tests */
1721 if (btrfs_test_is_dummy_root(root))
1722 return;
1723
1724 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1725 root->fs_info->delalloc_batch);
1726 spin_lock(&BTRFS_I(inode)->lock);
1727 BTRFS_I(inode)->delalloc_bytes += len;
1728 if (*bits & EXTENT_DEFRAG)
1729 BTRFS_I(inode)->defrag_bytes += len;
1730 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1731 &BTRFS_I(inode)->runtime_flags))
1732 btrfs_add_delalloc_inodes(root, inode);
1733 spin_unlock(&BTRFS_I(inode)->lock);
1734 }
1735 }
1736
1737 /*
1738 * extent_io.c clear_bit_hook, see set_bit_hook for why
1739 */
1740 static void btrfs_clear_bit_hook(struct inode *inode,
1741 struct extent_state *state,
1742 unsigned *bits)
1743 {
1744 u64 len = state->end + 1 - state->start;
1745 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1746 BTRFS_MAX_EXTENT_SIZE);
1747
1748 spin_lock(&BTRFS_I(inode)->lock);
1749 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1750 BTRFS_I(inode)->defrag_bytes -= len;
1751 spin_unlock(&BTRFS_I(inode)->lock);
1752
1753 /*
1754 * set_bit and clear bit hooks normally require _irqsave/restore
1755 * but in this case, we are only testing for the DELALLOC
1756 * bit, which is only set or cleared with irqs on
1757 */
1758 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1759 struct btrfs_root *root = BTRFS_I(inode)->root;
1760 bool do_list = !btrfs_is_free_space_inode(inode);
1761
1762 if (*bits & EXTENT_FIRST_DELALLOC) {
1763 *bits &= ~EXTENT_FIRST_DELALLOC;
1764 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1765 spin_lock(&BTRFS_I(inode)->lock);
1766 BTRFS_I(inode)->outstanding_extents -= num_extents;
1767 spin_unlock(&BTRFS_I(inode)->lock);
1768 }
1769
1770 /*
1771 * We don't reserve metadata space for space cache inodes so we
1772 * don't need to call dellalloc_release_metadata if there is an
1773 * error.
1774 */
1775 if (*bits & EXTENT_DO_ACCOUNTING &&
1776 root != root->fs_info->tree_root)
1777 btrfs_delalloc_release_metadata(inode, len);
1778
1779 /* For sanity tests. */
1780 if (btrfs_test_is_dummy_root(root))
1781 return;
1782
1783 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1784 && do_list && !(state->state & EXTENT_NORESERVE))
1785 btrfs_free_reserved_data_space_noquota(inode,
1786 state->start, len);
1787
1788 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1789 root->fs_info->delalloc_batch);
1790 spin_lock(&BTRFS_I(inode)->lock);
1791 BTRFS_I(inode)->delalloc_bytes -= len;
1792 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1793 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1794 &BTRFS_I(inode)->runtime_flags))
1795 btrfs_del_delalloc_inode(root, inode);
1796 spin_unlock(&BTRFS_I(inode)->lock);
1797 }
1798 }
1799
1800 /*
1801 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1802 * we don't create bios that span stripes or chunks
1803 */
1804 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1805 size_t size, struct bio *bio,
1806 unsigned long bio_flags)
1807 {
1808 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1809 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1810 u64 length = 0;
1811 u64 map_length;
1812 int ret;
1813
1814 if (bio_flags & EXTENT_BIO_COMPRESSED)
1815 return 0;
1816
1817 length = bio->bi_iter.bi_size;
1818 map_length = length;
1819 ret = btrfs_map_block(root->fs_info, rw, logical,
1820 &map_length, NULL, 0);
1821 /* Will always return 0 with map_multi == NULL */
1822 BUG_ON(ret < 0);
1823 if (map_length < length + size)
1824 return 1;
1825 return 0;
1826 }
1827
1828 /*
1829 * in order to insert checksums into the metadata in large chunks,
1830 * we wait until bio submission time. All the pages in the bio are
1831 * checksummed and sums are attached onto the ordered extent record.
1832 *
1833 * At IO completion time the cums attached on the ordered extent record
1834 * are inserted into the btree
1835 */
1836 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1837 struct bio *bio, int mirror_num,
1838 unsigned long bio_flags,
1839 u64 bio_offset)
1840 {
1841 struct btrfs_root *root = BTRFS_I(inode)->root;
1842 int ret = 0;
1843
1844 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1845 BUG_ON(ret); /* -ENOMEM */
1846 return 0;
1847 }
1848
1849 /*
1850 * in order to insert checksums into the metadata in large chunks,
1851 * we wait until bio submission time. All the pages in the bio are
1852 * checksummed and sums are attached onto the ordered extent record.
1853 *
1854 * At IO completion time the cums attached on the ordered extent record
1855 * are inserted into the btree
1856 */
1857 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1858 int mirror_num, unsigned long bio_flags,
1859 u64 bio_offset)
1860 {
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
1862 int ret;
1863
1864 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1865 if (ret) {
1866 bio->bi_error = ret;
1867 bio_endio(bio);
1868 }
1869 return ret;
1870 }
1871
1872 /*
1873 * extent_io.c submission hook. This does the right thing for csum calculation
1874 * on write, or reading the csums from the tree before a read
1875 */
1876 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1877 int mirror_num, unsigned long bio_flags,
1878 u64 bio_offset)
1879 {
1880 struct btrfs_root *root = BTRFS_I(inode)->root;
1881 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1882 int ret = 0;
1883 int skip_sum;
1884 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1885
1886 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1887
1888 if (btrfs_is_free_space_inode(inode))
1889 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1890
1891 if (!(rw & REQ_WRITE)) {
1892 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1893 if (ret)
1894 goto out;
1895
1896 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1897 ret = btrfs_submit_compressed_read(inode, bio,
1898 mirror_num,
1899 bio_flags);
1900 goto out;
1901 } else if (!skip_sum) {
1902 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1903 if (ret)
1904 goto out;
1905 }
1906 goto mapit;
1907 } else if (async && !skip_sum) {
1908 /* csum items have already been cloned */
1909 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1910 goto mapit;
1911 /* we're doing a write, do the async checksumming */
1912 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1913 inode, rw, bio, mirror_num,
1914 bio_flags, bio_offset,
1915 __btrfs_submit_bio_start,
1916 __btrfs_submit_bio_done);
1917 goto out;
1918 } else if (!skip_sum) {
1919 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1920 if (ret)
1921 goto out;
1922 }
1923
1924 mapit:
1925 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1926
1927 out:
1928 if (ret < 0) {
1929 bio->bi_error = ret;
1930 bio_endio(bio);
1931 }
1932 return ret;
1933 }
1934
1935 /*
1936 * given a list of ordered sums record them in the inode. This happens
1937 * at IO completion time based on sums calculated at bio submission time.
1938 */
1939 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1940 struct inode *inode, u64 file_offset,
1941 struct list_head *list)
1942 {
1943 struct btrfs_ordered_sum *sum;
1944
1945 list_for_each_entry(sum, list, list) {
1946 trans->adding_csums = 1;
1947 btrfs_csum_file_blocks(trans,
1948 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1949 trans->adding_csums = 0;
1950 }
1951 return 0;
1952 }
1953
1954 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1955 struct extent_state **cached_state)
1956 {
1957 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1958 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1959 cached_state, GFP_NOFS);
1960 }
1961
1962 /* see btrfs_writepage_start_hook for details on why this is required */
1963 struct btrfs_writepage_fixup {
1964 struct page *page;
1965 struct btrfs_work work;
1966 };
1967
1968 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1969 {
1970 struct btrfs_writepage_fixup *fixup;
1971 struct btrfs_ordered_extent *ordered;
1972 struct extent_state *cached_state = NULL;
1973 struct page *page;
1974 struct inode *inode;
1975 u64 page_start;
1976 u64 page_end;
1977 int ret;
1978
1979 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1980 page = fixup->page;
1981 again:
1982 lock_page(page);
1983 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1984 ClearPageChecked(page);
1985 goto out_page;
1986 }
1987
1988 inode = page->mapping->host;
1989 page_start = page_offset(page);
1990 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1991
1992 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1993 &cached_state);
1994
1995 /* already ordered? We're done */
1996 if (PagePrivate2(page))
1997 goto out;
1998
1999 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2000 if (ordered) {
2001 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2002 page_end, &cached_state, GFP_NOFS);
2003 unlock_page(page);
2004 btrfs_start_ordered_extent(inode, ordered, 1);
2005 btrfs_put_ordered_extent(ordered);
2006 goto again;
2007 }
2008
2009 ret = btrfs_delalloc_reserve_space(inode, page_start,
2010 PAGE_CACHE_SIZE);
2011 if (ret) {
2012 mapping_set_error(page->mapping, ret);
2013 end_extent_writepage(page, ret, page_start, page_end);
2014 ClearPageChecked(page);
2015 goto out;
2016 }
2017
2018 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2019 ClearPageChecked(page);
2020 set_page_dirty(page);
2021 out:
2022 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2023 &cached_state, GFP_NOFS);
2024 out_page:
2025 unlock_page(page);
2026 page_cache_release(page);
2027 kfree(fixup);
2028 }
2029
2030 /*
2031 * There are a few paths in the higher layers of the kernel that directly
2032 * set the page dirty bit without asking the filesystem if it is a
2033 * good idea. This causes problems because we want to make sure COW
2034 * properly happens and the data=ordered rules are followed.
2035 *
2036 * In our case any range that doesn't have the ORDERED bit set
2037 * hasn't been properly setup for IO. We kick off an async process
2038 * to fix it up. The async helper will wait for ordered extents, set
2039 * the delalloc bit and make it safe to write the page.
2040 */
2041 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2042 {
2043 struct inode *inode = page->mapping->host;
2044 struct btrfs_writepage_fixup *fixup;
2045 struct btrfs_root *root = BTRFS_I(inode)->root;
2046
2047 /* this page is properly in the ordered list */
2048 if (TestClearPagePrivate2(page))
2049 return 0;
2050
2051 if (PageChecked(page))
2052 return -EAGAIN;
2053
2054 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2055 if (!fixup)
2056 return -EAGAIN;
2057
2058 SetPageChecked(page);
2059 page_cache_get(page);
2060 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2061 btrfs_writepage_fixup_worker, NULL, NULL);
2062 fixup->page = page;
2063 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2064 return -EBUSY;
2065 }
2066
2067 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2068 struct inode *inode, u64 file_pos,
2069 u64 disk_bytenr, u64 disk_num_bytes,
2070 u64 num_bytes, u64 ram_bytes,
2071 u8 compression, u8 encryption,
2072 u16 other_encoding, int extent_type)
2073 {
2074 struct btrfs_root *root = BTRFS_I(inode)->root;
2075 struct btrfs_file_extent_item *fi;
2076 struct btrfs_path *path;
2077 struct extent_buffer *leaf;
2078 struct btrfs_key ins;
2079 int extent_inserted = 0;
2080 int ret;
2081
2082 path = btrfs_alloc_path();
2083 if (!path)
2084 return -ENOMEM;
2085
2086 /*
2087 * we may be replacing one extent in the tree with another.
2088 * The new extent is pinned in the extent map, and we don't want
2089 * to drop it from the cache until it is completely in the btree.
2090 *
2091 * So, tell btrfs_drop_extents to leave this extent in the cache.
2092 * the caller is expected to unpin it and allow it to be merged
2093 * with the others.
2094 */
2095 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2096 file_pos + num_bytes, NULL, 0,
2097 1, sizeof(*fi), &extent_inserted);
2098 if (ret)
2099 goto out;
2100
2101 if (!extent_inserted) {
2102 ins.objectid = btrfs_ino(inode);
2103 ins.offset = file_pos;
2104 ins.type = BTRFS_EXTENT_DATA_KEY;
2105
2106 path->leave_spinning = 1;
2107 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2108 sizeof(*fi));
2109 if (ret)
2110 goto out;
2111 }
2112 leaf = path->nodes[0];
2113 fi = btrfs_item_ptr(leaf, path->slots[0],
2114 struct btrfs_file_extent_item);
2115 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2116 btrfs_set_file_extent_type(leaf, fi, extent_type);
2117 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2118 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2119 btrfs_set_file_extent_offset(leaf, fi, 0);
2120 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2121 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2122 btrfs_set_file_extent_compression(leaf, fi, compression);
2123 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2124 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2125
2126 btrfs_mark_buffer_dirty(leaf);
2127 btrfs_release_path(path);
2128
2129 inode_add_bytes(inode, num_bytes);
2130
2131 ins.objectid = disk_bytenr;
2132 ins.offset = disk_num_bytes;
2133 ins.type = BTRFS_EXTENT_ITEM_KEY;
2134 ret = btrfs_alloc_reserved_file_extent(trans, root,
2135 root->root_key.objectid,
2136 btrfs_ino(inode), file_pos,
2137 ram_bytes, &ins);
2138 /*
2139 * Release the reserved range from inode dirty range map, as it is
2140 * already moved into delayed_ref_head
2141 */
2142 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2143 out:
2144 btrfs_free_path(path);
2145
2146 return ret;
2147 }
2148
2149 /* snapshot-aware defrag */
2150 struct sa_defrag_extent_backref {
2151 struct rb_node node;
2152 struct old_sa_defrag_extent *old;
2153 u64 root_id;
2154 u64 inum;
2155 u64 file_pos;
2156 u64 extent_offset;
2157 u64 num_bytes;
2158 u64 generation;
2159 };
2160
2161 struct old_sa_defrag_extent {
2162 struct list_head list;
2163 struct new_sa_defrag_extent *new;
2164
2165 u64 extent_offset;
2166 u64 bytenr;
2167 u64 offset;
2168 u64 len;
2169 int count;
2170 };
2171
2172 struct new_sa_defrag_extent {
2173 struct rb_root root;
2174 struct list_head head;
2175 struct btrfs_path *path;
2176 struct inode *inode;
2177 u64 file_pos;
2178 u64 len;
2179 u64 bytenr;
2180 u64 disk_len;
2181 u8 compress_type;
2182 };
2183
2184 static int backref_comp(struct sa_defrag_extent_backref *b1,
2185 struct sa_defrag_extent_backref *b2)
2186 {
2187 if (b1->root_id < b2->root_id)
2188 return -1;
2189 else if (b1->root_id > b2->root_id)
2190 return 1;
2191
2192 if (b1->inum < b2->inum)
2193 return -1;
2194 else if (b1->inum > b2->inum)
2195 return 1;
2196
2197 if (b1->file_pos < b2->file_pos)
2198 return -1;
2199 else if (b1->file_pos > b2->file_pos)
2200 return 1;
2201
2202 /*
2203 * [------------------------------] ===> (a range of space)
2204 * |<--->| |<---->| =============> (fs/file tree A)
2205 * |<---------------------------->| ===> (fs/file tree B)
2206 *
2207 * A range of space can refer to two file extents in one tree while
2208 * refer to only one file extent in another tree.
2209 *
2210 * So we may process a disk offset more than one time(two extents in A)
2211 * and locate at the same extent(one extent in B), then insert two same
2212 * backrefs(both refer to the extent in B).
2213 */
2214 return 0;
2215 }
2216
2217 static void backref_insert(struct rb_root *root,
2218 struct sa_defrag_extent_backref *backref)
2219 {
2220 struct rb_node **p = &root->rb_node;
2221 struct rb_node *parent = NULL;
2222 struct sa_defrag_extent_backref *entry;
2223 int ret;
2224
2225 while (*p) {
2226 parent = *p;
2227 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2228
2229 ret = backref_comp(backref, entry);
2230 if (ret < 0)
2231 p = &(*p)->rb_left;
2232 else
2233 p = &(*p)->rb_right;
2234 }
2235
2236 rb_link_node(&backref->node, parent, p);
2237 rb_insert_color(&backref->node, root);
2238 }
2239
2240 /*
2241 * Note the backref might has changed, and in this case we just return 0.
2242 */
2243 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2244 void *ctx)
2245 {
2246 struct btrfs_file_extent_item *extent;
2247 struct btrfs_fs_info *fs_info;
2248 struct old_sa_defrag_extent *old = ctx;
2249 struct new_sa_defrag_extent *new = old->new;
2250 struct btrfs_path *path = new->path;
2251 struct btrfs_key key;
2252 struct btrfs_root *root;
2253 struct sa_defrag_extent_backref *backref;
2254 struct extent_buffer *leaf;
2255 struct inode *inode = new->inode;
2256 int slot;
2257 int ret;
2258 u64 extent_offset;
2259 u64 num_bytes;
2260
2261 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2262 inum == btrfs_ino(inode))
2263 return 0;
2264
2265 key.objectid = root_id;
2266 key.type = BTRFS_ROOT_ITEM_KEY;
2267 key.offset = (u64)-1;
2268
2269 fs_info = BTRFS_I(inode)->root->fs_info;
2270 root = btrfs_read_fs_root_no_name(fs_info, &key);
2271 if (IS_ERR(root)) {
2272 if (PTR_ERR(root) == -ENOENT)
2273 return 0;
2274 WARN_ON(1);
2275 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2276 inum, offset, root_id);
2277 return PTR_ERR(root);
2278 }
2279
2280 key.objectid = inum;
2281 key.type = BTRFS_EXTENT_DATA_KEY;
2282 if (offset > (u64)-1 << 32)
2283 key.offset = 0;
2284 else
2285 key.offset = offset;
2286
2287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2288 if (WARN_ON(ret < 0))
2289 return ret;
2290 ret = 0;
2291
2292 while (1) {
2293 cond_resched();
2294
2295 leaf = path->nodes[0];
2296 slot = path->slots[0];
2297
2298 if (slot >= btrfs_header_nritems(leaf)) {
2299 ret = btrfs_next_leaf(root, path);
2300 if (ret < 0) {
2301 goto out;
2302 } else if (ret > 0) {
2303 ret = 0;
2304 goto out;
2305 }
2306 continue;
2307 }
2308
2309 path->slots[0]++;
2310
2311 btrfs_item_key_to_cpu(leaf, &key, slot);
2312
2313 if (key.objectid > inum)
2314 goto out;
2315
2316 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2317 continue;
2318
2319 extent = btrfs_item_ptr(leaf, slot,
2320 struct btrfs_file_extent_item);
2321
2322 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2323 continue;
2324
2325 /*
2326 * 'offset' refers to the exact key.offset,
2327 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2328 * (key.offset - extent_offset).
2329 */
2330 if (key.offset != offset)
2331 continue;
2332
2333 extent_offset = btrfs_file_extent_offset(leaf, extent);
2334 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2335
2336 if (extent_offset >= old->extent_offset + old->offset +
2337 old->len || extent_offset + num_bytes <=
2338 old->extent_offset + old->offset)
2339 continue;
2340 break;
2341 }
2342
2343 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2344 if (!backref) {
2345 ret = -ENOENT;
2346 goto out;
2347 }
2348
2349 backref->root_id = root_id;
2350 backref->inum = inum;
2351 backref->file_pos = offset;
2352 backref->num_bytes = num_bytes;
2353 backref->extent_offset = extent_offset;
2354 backref->generation = btrfs_file_extent_generation(leaf, extent);
2355 backref->old = old;
2356 backref_insert(&new->root, backref);
2357 old->count++;
2358 out:
2359 btrfs_release_path(path);
2360 WARN_ON(ret);
2361 return ret;
2362 }
2363
2364 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2365 struct new_sa_defrag_extent *new)
2366 {
2367 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2368 struct old_sa_defrag_extent *old, *tmp;
2369 int ret;
2370
2371 new->path = path;
2372
2373 list_for_each_entry_safe(old, tmp, &new->head, list) {
2374 ret = iterate_inodes_from_logical(old->bytenr +
2375 old->extent_offset, fs_info,
2376 path, record_one_backref,
2377 old);
2378 if (ret < 0 && ret != -ENOENT)
2379 return false;
2380
2381 /* no backref to be processed for this extent */
2382 if (!old->count) {
2383 list_del(&old->list);
2384 kfree(old);
2385 }
2386 }
2387
2388 if (list_empty(&new->head))
2389 return false;
2390
2391 return true;
2392 }
2393
2394 static int relink_is_mergable(struct extent_buffer *leaf,
2395 struct btrfs_file_extent_item *fi,
2396 struct new_sa_defrag_extent *new)
2397 {
2398 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2399 return 0;
2400
2401 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2402 return 0;
2403
2404 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2405 return 0;
2406
2407 if (btrfs_file_extent_encryption(leaf, fi) ||
2408 btrfs_file_extent_other_encoding(leaf, fi))
2409 return 0;
2410
2411 return 1;
2412 }
2413
2414 /*
2415 * Note the backref might has changed, and in this case we just return 0.
2416 */
2417 static noinline int relink_extent_backref(struct btrfs_path *path,
2418 struct sa_defrag_extent_backref *prev,
2419 struct sa_defrag_extent_backref *backref)
2420 {
2421 struct btrfs_file_extent_item *extent;
2422 struct btrfs_file_extent_item *item;
2423 struct btrfs_ordered_extent *ordered;
2424 struct btrfs_trans_handle *trans;
2425 struct btrfs_fs_info *fs_info;
2426 struct btrfs_root *root;
2427 struct btrfs_key key;
2428 struct extent_buffer *leaf;
2429 struct old_sa_defrag_extent *old = backref->old;
2430 struct new_sa_defrag_extent *new = old->new;
2431 struct inode *src_inode = new->inode;
2432 struct inode *inode;
2433 struct extent_state *cached = NULL;
2434 int ret = 0;
2435 u64 start;
2436 u64 len;
2437 u64 lock_start;
2438 u64 lock_end;
2439 bool merge = false;
2440 int index;
2441
2442 if (prev && prev->root_id == backref->root_id &&
2443 prev->inum == backref->inum &&
2444 prev->file_pos + prev->num_bytes == backref->file_pos)
2445 merge = true;
2446
2447 /* step 1: get root */
2448 key.objectid = backref->root_id;
2449 key.type = BTRFS_ROOT_ITEM_KEY;
2450 key.offset = (u64)-1;
2451
2452 fs_info = BTRFS_I(src_inode)->root->fs_info;
2453 index = srcu_read_lock(&fs_info->subvol_srcu);
2454
2455 root = btrfs_read_fs_root_no_name(fs_info, &key);
2456 if (IS_ERR(root)) {
2457 srcu_read_unlock(&fs_info->subvol_srcu, index);
2458 if (PTR_ERR(root) == -ENOENT)
2459 return 0;
2460 return PTR_ERR(root);
2461 }
2462
2463 if (btrfs_root_readonly(root)) {
2464 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465 return 0;
2466 }
2467
2468 /* step 2: get inode */
2469 key.objectid = backref->inum;
2470 key.type = BTRFS_INODE_ITEM_KEY;
2471 key.offset = 0;
2472
2473 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2474 if (IS_ERR(inode)) {
2475 srcu_read_unlock(&fs_info->subvol_srcu, index);
2476 return 0;
2477 }
2478
2479 srcu_read_unlock(&fs_info->subvol_srcu, index);
2480
2481 /* step 3: relink backref */
2482 lock_start = backref->file_pos;
2483 lock_end = backref->file_pos + backref->num_bytes - 1;
2484 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2485 0, &cached);
2486
2487 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2488 if (ordered) {
2489 btrfs_put_ordered_extent(ordered);
2490 goto out_unlock;
2491 }
2492
2493 trans = btrfs_join_transaction(root);
2494 if (IS_ERR(trans)) {
2495 ret = PTR_ERR(trans);
2496 goto out_unlock;
2497 }
2498
2499 key.objectid = backref->inum;
2500 key.type = BTRFS_EXTENT_DATA_KEY;
2501 key.offset = backref->file_pos;
2502
2503 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2504 if (ret < 0) {
2505 goto out_free_path;
2506 } else if (ret > 0) {
2507 ret = 0;
2508 goto out_free_path;
2509 }
2510
2511 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2512 struct btrfs_file_extent_item);
2513
2514 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2515 backref->generation)
2516 goto out_free_path;
2517
2518 btrfs_release_path(path);
2519
2520 start = backref->file_pos;
2521 if (backref->extent_offset < old->extent_offset + old->offset)
2522 start += old->extent_offset + old->offset -
2523 backref->extent_offset;
2524
2525 len = min(backref->extent_offset + backref->num_bytes,
2526 old->extent_offset + old->offset + old->len);
2527 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2528
2529 ret = btrfs_drop_extents(trans, root, inode, start,
2530 start + len, 1);
2531 if (ret)
2532 goto out_free_path;
2533 again:
2534 key.objectid = btrfs_ino(inode);
2535 key.type = BTRFS_EXTENT_DATA_KEY;
2536 key.offset = start;
2537
2538 path->leave_spinning = 1;
2539 if (merge) {
2540 struct btrfs_file_extent_item *fi;
2541 u64 extent_len;
2542 struct btrfs_key found_key;
2543
2544 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2545 if (ret < 0)
2546 goto out_free_path;
2547
2548 path->slots[0]--;
2549 leaf = path->nodes[0];
2550 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2551
2552 fi = btrfs_item_ptr(leaf, path->slots[0],
2553 struct btrfs_file_extent_item);
2554 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2555
2556 if (extent_len + found_key.offset == start &&
2557 relink_is_mergable(leaf, fi, new)) {
2558 btrfs_set_file_extent_num_bytes(leaf, fi,
2559 extent_len + len);
2560 btrfs_mark_buffer_dirty(leaf);
2561 inode_add_bytes(inode, len);
2562
2563 ret = 1;
2564 goto out_free_path;
2565 } else {
2566 merge = false;
2567 btrfs_release_path(path);
2568 goto again;
2569 }
2570 }
2571
2572 ret = btrfs_insert_empty_item(trans, root, path, &key,
2573 sizeof(*extent));
2574 if (ret) {
2575 btrfs_abort_transaction(trans, root, ret);
2576 goto out_free_path;
2577 }
2578
2579 leaf = path->nodes[0];
2580 item = btrfs_item_ptr(leaf, path->slots[0],
2581 struct btrfs_file_extent_item);
2582 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2583 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2584 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2585 btrfs_set_file_extent_num_bytes(leaf, item, len);
2586 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2587 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2588 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2589 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2590 btrfs_set_file_extent_encryption(leaf, item, 0);
2591 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2592
2593 btrfs_mark_buffer_dirty(leaf);
2594 inode_add_bytes(inode, len);
2595 btrfs_release_path(path);
2596
2597 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2598 new->disk_len, 0,
2599 backref->root_id, backref->inum,
2600 new->file_pos); /* start - extent_offset */
2601 if (ret) {
2602 btrfs_abort_transaction(trans, root, ret);
2603 goto out_free_path;
2604 }
2605
2606 ret = 1;
2607 out_free_path:
2608 btrfs_release_path(path);
2609 path->leave_spinning = 0;
2610 btrfs_end_transaction(trans, root);
2611 out_unlock:
2612 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2613 &cached, GFP_NOFS);
2614 iput(inode);
2615 return ret;
2616 }
2617
2618 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2619 {
2620 struct old_sa_defrag_extent *old, *tmp;
2621
2622 if (!new)
2623 return;
2624
2625 list_for_each_entry_safe(old, tmp, &new->head, list) {
2626 kfree(old);
2627 }
2628 kfree(new);
2629 }
2630
2631 static void relink_file_extents(struct new_sa_defrag_extent *new)
2632 {
2633 struct btrfs_path *path;
2634 struct sa_defrag_extent_backref *backref;
2635 struct sa_defrag_extent_backref *prev = NULL;
2636 struct inode *inode;
2637 struct btrfs_root *root;
2638 struct rb_node *node;
2639 int ret;
2640
2641 inode = new->inode;
2642 root = BTRFS_I(inode)->root;
2643
2644 path = btrfs_alloc_path();
2645 if (!path)
2646 return;
2647
2648 if (!record_extent_backrefs(path, new)) {
2649 btrfs_free_path(path);
2650 goto out;
2651 }
2652 btrfs_release_path(path);
2653
2654 while (1) {
2655 node = rb_first(&new->root);
2656 if (!node)
2657 break;
2658 rb_erase(node, &new->root);
2659
2660 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2661
2662 ret = relink_extent_backref(path, prev, backref);
2663 WARN_ON(ret < 0);
2664
2665 kfree(prev);
2666
2667 if (ret == 1)
2668 prev = backref;
2669 else
2670 prev = NULL;
2671 cond_resched();
2672 }
2673 kfree(prev);
2674
2675 btrfs_free_path(path);
2676 out:
2677 free_sa_defrag_extent(new);
2678
2679 atomic_dec(&root->fs_info->defrag_running);
2680 wake_up(&root->fs_info->transaction_wait);
2681 }
2682
2683 static struct new_sa_defrag_extent *
2684 record_old_file_extents(struct inode *inode,
2685 struct btrfs_ordered_extent *ordered)
2686 {
2687 struct btrfs_root *root = BTRFS_I(inode)->root;
2688 struct btrfs_path *path;
2689 struct btrfs_key key;
2690 struct old_sa_defrag_extent *old;
2691 struct new_sa_defrag_extent *new;
2692 int ret;
2693
2694 new = kmalloc(sizeof(*new), GFP_NOFS);
2695 if (!new)
2696 return NULL;
2697
2698 new->inode = inode;
2699 new->file_pos = ordered->file_offset;
2700 new->len = ordered->len;
2701 new->bytenr = ordered->start;
2702 new->disk_len = ordered->disk_len;
2703 new->compress_type = ordered->compress_type;
2704 new->root = RB_ROOT;
2705 INIT_LIST_HEAD(&new->head);
2706
2707 path = btrfs_alloc_path();
2708 if (!path)
2709 goto out_kfree;
2710
2711 key.objectid = btrfs_ino(inode);
2712 key.type = BTRFS_EXTENT_DATA_KEY;
2713 key.offset = new->file_pos;
2714
2715 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2716 if (ret < 0)
2717 goto out_free_path;
2718 if (ret > 0 && path->slots[0] > 0)
2719 path->slots[0]--;
2720
2721 /* find out all the old extents for the file range */
2722 while (1) {
2723 struct btrfs_file_extent_item *extent;
2724 struct extent_buffer *l;
2725 int slot;
2726 u64 num_bytes;
2727 u64 offset;
2728 u64 end;
2729 u64 disk_bytenr;
2730 u64 extent_offset;
2731
2732 l = path->nodes[0];
2733 slot = path->slots[0];
2734
2735 if (slot >= btrfs_header_nritems(l)) {
2736 ret = btrfs_next_leaf(root, path);
2737 if (ret < 0)
2738 goto out_free_path;
2739 else if (ret > 0)
2740 break;
2741 continue;
2742 }
2743
2744 btrfs_item_key_to_cpu(l, &key, slot);
2745
2746 if (key.objectid != btrfs_ino(inode))
2747 break;
2748 if (key.type != BTRFS_EXTENT_DATA_KEY)
2749 break;
2750 if (key.offset >= new->file_pos + new->len)
2751 break;
2752
2753 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2754
2755 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2756 if (key.offset + num_bytes < new->file_pos)
2757 goto next;
2758
2759 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2760 if (!disk_bytenr)
2761 goto next;
2762
2763 extent_offset = btrfs_file_extent_offset(l, extent);
2764
2765 old = kmalloc(sizeof(*old), GFP_NOFS);
2766 if (!old)
2767 goto out_free_path;
2768
2769 offset = max(new->file_pos, key.offset);
2770 end = min(new->file_pos + new->len, key.offset + num_bytes);
2771
2772 old->bytenr = disk_bytenr;
2773 old->extent_offset = extent_offset;
2774 old->offset = offset - key.offset;
2775 old->len = end - offset;
2776 old->new = new;
2777 old->count = 0;
2778 list_add_tail(&old->list, &new->head);
2779 next:
2780 path->slots[0]++;
2781 cond_resched();
2782 }
2783
2784 btrfs_free_path(path);
2785 atomic_inc(&root->fs_info->defrag_running);
2786
2787 return new;
2788
2789 out_free_path:
2790 btrfs_free_path(path);
2791 out_kfree:
2792 free_sa_defrag_extent(new);
2793 return NULL;
2794 }
2795
2796 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2797 u64 start, u64 len)
2798 {
2799 struct btrfs_block_group_cache *cache;
2800
2801 cache = btrfs_lookup_block_group(root->fs_info, start);
2802 ASSERT(cache);
2803
2804 spin_lock(&cache->lock);
2805 cache->delalloc_bytes -= len;
2806 spin_unlock(&cache->lock);
2807
2808 btrfs_put_block_group(cache);
2809 }
2810
2811 /* as ordered data IO finishes, this gets called so we can finish
2812 * an ordered extent if the range of bytes in the file it covers are
2813 * fully written.
2814 */
2815 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2816 {
2817 struct inode *inode = ordered_extent->inode;
2818 struct btrfs_root *root = BTRFS_I(inode)->root;
2819 struct btrfs_trans_handle *trans = NULL;
2820 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2821 struct extent_state *cached_state = NULL;
2822 struct new_sa_defrag_extent *new = NULL;
2823 int compress_type = 0;
2824 int ret = 0;
2825 u64 logical_len = ordered_extent->len;
2826 bool nolock;
2827 bool truncated = false;
2828
2829 nolock = btrfs_is_free_space_inode(inode);
2830
2831 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2832 ret = -EIO;
2833 goto out;
2834 }
2835
2836 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2837 ordered_extent->file_offset +
2838 ordered_extent->len - 1);
2839
2840 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2841 truncated = true;
2842 logical_len = ordered_extent->truncated_len;
2843 /* Truncated the entire extent, don't bother adding */
2844 if (!logical_len)
2845 goto out;
2846 }
2847
2848 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2849 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2850
2851 /*
2852 * For mwrite(mmap + memset to write) case, we still reserve
2853 * space for NOCOW range.
2854 * As NOCOW won't cause a new delayed ref, just free the space
2855 */
2856 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2857 ordered_extent->len);
2858 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2859 if (nolock)
2860 trans = btrfs_join_transaction_nolock(root);
2861 else
2862 trans = btrfs_join_transaction(root);
2863 if (IS_ERR(trans)) {
2864 ret = PTR_ERR(trans);
2865 trans = NULL;
2866 goto out;
2867 }
2868 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2869 ret = btrfs_update_inode_fallback(trans, root, inode);
2870 if (ret) /* -ENOMEM or corruption */
2871 btrfs_abort_transaction(trans, root, ret);
2872 goto out;
2873 }
2874
2875 lock_extent_bits(io_tree, ordered_extent->file_offset,
2876 ordered_extent->file_offset + ordered_extent->len - 1,
2877 0, &cached_state);
2878
2879 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2880 ordered_extent->file_offset + ordered_extent->len - 1,
2881 EXTENT_DEFRAG, 1, cached_state);
2882 if (ret) {
2883 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2884 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2885 /* the inode is shared */
2886 new = record_old_file_extents(inode, ordered_extent);
2887
2888 clear_extent_bit(io_tree, ordered_extent->file_offset,
2889 ordered_extent->file_offset + ordered_extent->len - 1,
2890 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2891 }
2892
2893 if (nolock)
2894 trans = btrfs_join_transaction_nolock(root);
2895 else
2896 trans = btrfs_join_transaction(root);
2897 if (IS_ERR(trans)) {
2898 ret = PTR_ERR(trans);
2899 trans = NULL;
2900 goto out_unlock;
2901 }
2902
2903 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2904
2905 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2906 compress_type = ordered_extent->compress_type;
2907 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2908 BUG_ON(compress_type);
2909 ret = btrfs_mark_extent_written(trans, inode,
2910 ordered_extent->file_offset,
2911 ordered_extent->file_offset +
2912 logical_len);
2913 } else {
2914 BUG_ON(root == root->fs_info->tree_root);
2915 ret = insert_reserved_file_extent(trans, inode,
2916 ordered_extent->file_offset,
2917 ordered_extent->start,
2918 ordered_extent->disk_len,
2919 logical_len, logical_len,
2920 compress_type, 0, 0,
2921 BTRFS_FILE_EXTENT_REG);
2922 if (!ret)
2923 btrfs_release_delalloc_bytes(root,
2924 ordered_extent->start,
2925 ordered_extent->disk_len);
2926 }
2927 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2928 ordered_extent->file_offset, ordered_extent->len,
2929 trans->transid);
2930 if (ret < 0) {
2931 btrfs_abort_transaction(trans, root, ret);
2932 goto out_unlock;
2933 }
2934
2935 add_pending_csums(trans, inode, ordered_extent->file_offset,
2936 &ordered_extent->list);
2937
2938 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2939 ret = btrfs_update_inode_fallback(trans, root, inode);
2940 if (ret) { /* -ENOMEM or corruption */
2941 btrfs_abort_transaction(trans, root, ret);
2942 goto out_unlock;
2943 }
2944 ret = 0;
2945 out_unlock:
2946 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2947 ordered_extent->file_offset +
2948 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2949 out:
2950 if (root != root->fs_info->tree_root)
2951 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2952 if (trans)
2953 btrfs_end_transaction(trans, root);
2954
2955 if (ret || truncated) {
2956 u64 start, end;
2957
2958 if (truncated)
2959 start = ordered_extent->file_offset + logical_len;
2960 else
2961 start = ordered_extent->file_offset;
2962 end = ordered_extent->file_offset + ordered_extent->len - 1;
2963 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2964
2965 /* Drop the cache for the part of the extent we didn't write. */
2966 btrfs_drop_extent_cache(inode, start, end, 0);
2967
2968 /*
2969 * If the ordered extent had an IOERR or something else went
2970 * wrong we need to return the space for this ordered extent
2971 * back to the allocator. We only free the extent in the
2972 * truncated case if we didn't write out the extent at all.
2973 */
2974 if ((ret || !logical_len) &&
2975 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2976 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2977 btrfs_free_reserved_extent(root, ordered_extent->start,
2978 ordered_extent->disk_len, 1);
2979 }
2980
2981
2982 /*
2983 * This needs to be done to make sure anybody waiting knows we are done
2984 * updating everything for this ordered extent.
2985 */
2986 btrfs_remove_ordered_extent(inode, ordered_extent);
2987
2988 /* for snapshot-aware defrag */
2989 if (new) {
2990 if (ret) {
2991 free_sa_defrag_extent(new);
2992 atomic_dec(&root->fs_info->defrag_running);
2993 } else {
2994 relink_file_extents(new);
2995 }
2996 }
2997
2998 /* once for us */
2999 btrfs_put_ordered_extent(ordered_extent);
3000 /* once for the tree */
3001 btrfs_put_ordered_extent(ordered_extent);
3002
3003 return ret;
3004 }
3005
3006 static void finish_ordered_fn(struct btrfs_work *work)
3007 {
3008 struct btrfs_ordered_extent *ordered_extent;
3009 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3010 btrfs_finish_ordered_io(ordered_extent);
3011 }
3012
3013 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3014 struct extent_state *state, int uptodate)
3015 {
3016 struct inode *inode = page->mapping->host;
3017 struct btrfs_root *root = BTRFS_I(inode)->root;
3018 struct btrfs_ordered_extent *ordered_extent = NULL;
3019 struct btrfs_workqueue *wq;
3020 btrfs_work_func_t func;
3021
3022 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3023
3024 ClearPagePrivate2(page);
3025 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3026 end - start + 1, uptodate))
3027 return 0;
3028
3029 if (btrfs_is_free_space_inode(inode)) {
3030 wq = root->fs_info->endio_freespace_worker;
3031 func = btrfs_freespace_write_helper;
3032 } else {
3033 wq = root->fs_info->endio_write_workers;
3034 func = btrfs_endio_write_helper;
3035 }
3036
3037 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3038 NULL);
3039 btrfs_queue_work(wq, &ordered_extent->work);
3040
3041 return 0;
3042 }
3043
3044 static int __readpage_endio_check(struct inode *inode,
3045 struct btrfs_io_bio *io_bio,
3046 int icsum, struct page *page,
3047 int pgoff, u64 start, size_t len)
3048 {
3049 char *kaddr;
3050 u32 csum_expected;
3051 u32 csum = ~(u32)0;
3052
3053 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3054
3055 kaddr = kmap_atomic(page);
3056 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3057 btrfs_csum_final(csum, (char *)&csum);
3058 if (csum != csum_expected)
3059 goto zeroit;
3060
3061 kunmap_atomic(kaddr);
3062 return 0;
3063 zeroit:
3064 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3065 "csum failed ino %llu off %llu csum %u expected csum %u",
3066 btrfs_ino(inode), start, csum, csum_expected);
3067 memset(kaddr + pgoff, 1, len);
3068 flush_dcache_page(page);
3069 kunmap_atomic(kaddr);
3070 if (csum_expected == 0)
3071 return 0;
3072 return -EIO;
3073 }
3074
3075 /*
3076 * when reads are done, we need to check csums to verify the data is correct
3077 * if there's a match, we allow the bio to finish. If not, the code in
3078 * extent_io.c will try to find good copies for us.
3079 */
3080 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3081 u64 phy_offset, struct page *page,
3082 u64 start, u64 end, int mirror)
3083 {
3084 size_t offset = start - page_offset(page);
3085 struct inode *inode = page->mapping->host;
3086 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3087 struct btrfs_root *root = BTRFS_I(inode)->root;
3088
3089 if (PageChecked(page)) {
3090 ClearPageChecked(page);
3091 return 0;
3092 }
3093
3094 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3095 return 0;
3096
3097 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3098 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3099 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3100 GFP_NOFS);
3101 return 0;
3102 }
3103
3104 phy_offset >>= inode->i_sb->s_blocksize_bits;
3105 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3106 start, (size_t)(end - start + 1));
3107 }
3108
3109 struct delayed_iput {
3110 struct list_head list;
3111 struct inode *inode;
3112 };
3113
3114 /* JDM: If this is fs-wide, why can't we add a pointer to
3115 * btrfs_inode instead and avoid the allocation? */
3116 void btrfs_add_delayed_iput(struct inode *inode)
3117 {
3118 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3119 struct delayed_iput *delayed;
3120
3121 if (atomic_add_unless(&inode->i_count, -1, 1))
3122 return;
3123
3124 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3125 delayed->inode = inode;
3126
3127 spin_lock(&fs_info->delayed_iput_lock);
3128 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3129 spin_unlock(&fs_info->delayed_iput_lock);
3130 }
3131
3132 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3133 {
3134 LIST_HEAD(list);
3135 struct btrfs_fs_info *fs_info = root->fs_info;
3136 struct delayed_iput *delayed;
3137 int empty;
3138
3139 spin_lock(&fs_info->delayed_iput_lock);
3140 empty = list_empty(&fs_info->delayed_iputs);
3141 spin_unlock(&fs_info->delayed_iput_lock);
3142 if (empty)
3143 return;
3144
3145 down_read(&fs_info->delayed_iput_sem);
3146
3147 spin_lock(&fs_info->delayed_iput_lock);
3148 list_splice_init(&fs_info->delayed_iputs, &list);
3149 spin_unlock(&fs_info->delayed_iput_lock);
3150
3151 while (!list_empty(&list)) {
3152 delayed = list_entry(list.next, struct delayed_iput, list);
3153 list_del(&delayed->list);
3154 iput(delayed->inode);
3155 kfree(delayed);
3156 }
3157
3158 up_read(&root->fs_info->delayed_iput_sem);
3159 }
3160
3161 /*
3162 * This is called in transaction commit time. If there are no orphan
3163 * files in the subvolume, it removes orphan item and frees block_rsv
3164 * structure.
3165 */
3166 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3167 struct btrfs_root *root)
3168 {
3169 struct btrfs_block_rsv *block_rsv;
3170 int ret;
3171
3172 if (atomic_read(&root->orphan_inodes) ||
3173 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3174 return;
3175
3176 spin_lock(&root->orphan_lock);
3177 if (atomic_read(&root->orphan_inodes)) {
3178 spin_unlock(&root->orphan_lock);
3179 return;
3180 }
3181
3182 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3183 spin_unlock(&root->orphan_lock);
3184 return;
3185 }
3186
3187 block_rsv = root->orphan_block_rsv;
3188 root->orphan_block_rsv = NULL;
3189 spin_unlock(&root->orphan_lock);
3190
3191 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3192 btrfs_root_refs(&root->root_item) > 0) {
3193 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3194 root->root_key.objectid);
3195 if (ret)
3196 btrfs_abort_transaction(trans, root, ret);
3197 else
3198 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3199 &root->state);
3200 }
3201
3202 if (block_rsv) {
3203 WARN_ON(block_rsv->size > 0);
3204 btrfs_free_block_rsv(root, block_rsv);
3205 }
3206 }
3207
3208 /*
3209 * This creates an orphan entry for the given inode in case something goes
3210 * wrong in the middle of an unlink/truncate.
3211 *
3212 * NOTE: caller of this function should reserve 5 units of metadata for
3213 * this function.
3214 */
3215 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3216 {
3217 struct btrfs_root *root = BTRFS_I(inode)->root;
3218 struct btrfs_block_rsv *block_rsv = NULL;
3219 int reserve = 0;
3220 int insert = 0;
3221 int ret;
3222
3223 if (!root->orphan_block_rsv) {
3224 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3225 if (!block_rsv)
3226 return -ENOMEM;
3227 }
3228
3229 spin_lock(&root->orphan_lock);
3230 if (!root->orphan_block_rsv) {
3231 root->orphan_block_rsv = block_rsv;
3232 } else if (block_rsv) {
3233 btrfs_free_block_rsv(root, block_rsv);
3234 block_rsv = NULL;
3235 }
3236
3237 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3238 &BTRFS_I(inode)->runtime_flags)) {
3239 #if 0
3240 /*
3241 * For proper ENOSPC handling, we should do orphan
3242 * cleanup when mounting. But this introduces backward
3243 * compatibility issue.
3244 */
3245 if (!xchg(&root->orphan_item_inserted, 1))
3246 insert = 2;
3247 else
3248 insert = 1;
3249 #endif
3250 insert = 1;
3251 atomic_inc(&root->orphan_inodes);
3252 }
3253
3254 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3255 &BTRFS_I(inode)->runtime_flags))
3256 reserve = 1;
3257 spin_unlock(&root->orphan_lock);
3258
3259 /* grab metadata reservation from transaction handle */
3260 if (reserve) {
3261 ret = btrfs_orphan_reserve_metadata(trans, inode);
3262 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3263 }
3264
3265 /* insert an orphan item to track this unlinked/truncated file */
3266 if (insert >= 1) {
3267 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3268 if (ret) {
3269 atomic_dec(&root->orphan_inodes);
3270 if (reserve) {
3271 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3272 &BTRFS_I(inode)->runtime_flags);
3273 btrfs_orphan_release_metadata(inode);
3274 }
3275 if (ret != -EEXIST) {
3276 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3277 &BTRFS_I(inode)->runtime_flags);
3278 btrfs_abort_transaction(trans, root, ret);
3279 return ret;
3280 }
3281 }
3282 ret = 0;
3283 }
3284
3285 /* insert an orphan item to track subvolume contains orphan files */
3286 if (insert >= 2) {
3287 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3288 root->root_key.objectid);
3289 if (ret && ret != -EEXIST) {
3290 btrfs_abort_transaction(trans, root, ret);
3291 return ret;
3292 }
3293 }
3294 return 0;
3295 }
3296
3297 /*
3298 * We have done the truncate/delete so we can go ahead and remove the orphan
3299 * item for this particular inode.
3300 */
3301 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3302 struct inode *inode)
3303 {
3304 struct btrfs_root *root = BTRFS_I(inode)->root;
3305 int delete_item = 0;
3306 int release_rsv = 0;
3307 int ret = 0;
3308
3309 spin_lock(&root->orphan_lock);
3310 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3311 &BTRFS_I(inode)->runtime_flags))
3312 delete_item = 1;
3313
3314 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3315 &BTRFS_I(inode)->runtime_flags))
3316 release_rsv = 1;
3317 spin_unlock(&root->orphan_lock);
3318
3319 if (delete_item) {
3320 atomic_dec(&root->orphan_inodes);
3321 if (trans)
3322 ret = btrfs_del_orphan_item(trans, root,
3323 btrfs_ino(inode));
3324 }
3325
3326 if (release_rsv)
3327 btrfs_orphan_release_metadata(inode);
3328
3329 return ret;
3330 }
3331
3332 /*
3333 * this cleans up any orphans that may be left on the list from the last use
3334 * of this root.
3335 */
3336 int btrfs_orphan_cleanup(struct btrfs_root *root)
3337 {
3338 struct btrfs_path *path;
3339 struct extent_buffer *leaf;
3340 struct btrfs_key key, found_key;
3341 struct btrfs_trans_handle *trans;
3342 struct inode *inode;
3343 u64 last_objectid = 0;
3344 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3345
3346 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3347 return 0;
3348
3349 path = btrfs_alloc_path();
3350 if (!path) {
3351 ret = -ENOMEM;
3352 goto out;
3353 }
3354 path->reada = -1;
3355
3356 key.objectid = BTRFS_ORPHAN_OBJECTID;
3357 key.type = BTRFS_ORPHAN_ITEM_KEY;
3358 key.offset = (u64)-1;
3359
3360 while (1) {
3361 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3362 if (ret < 0)
3363 goto out;
3364
3365 /*
3366 * if ret == 0 means we found what we were searching for, which
3367 * is weird, but possible, so only screw with path if we didn't
3368 * find the key and see if we have stuff that matches
3369 */
3370 if (ret > 0) {
3371 ret = 0;
3372 if (path->slots[0] == 0)
3373 break;
3374 path->slots[0]--;
3375 }
3376
3377 /* pull out the item */
3378 leaf = path->nodes[0];
3379 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3380
3381 /* make sure the item matches what we want */
3382 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3383 break;
3384 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3385 break;
3386
3387 /* release the path since we're done with it */
3388 btrfs_release_path(path);
3389
3390 /*
3391 * this is where we are basically btrfs_lookup, without the
3392 * crossing root thing. we store the inode number in the
3393 * offset of the orphan item.
3394 */
3395
3396 if (found_key.offset == last_objectid) {
3397 btrfs_err(root->fs_info,
3398 "Error removing orphan entry, stopping orphan cleanup");
3399 ret = -EINVAL;
3400 goto out;
3401 }
3402
3403 last_objectid = found_key.offset;
3404
3405 found_key.objectid = found_key.offset;
3406 found_key.type = BTRFS_INODE_ITEM_KEY;
3407 found_key.offset = 0;
3408 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3409 ret = PTR_ERR_OR_ZERO(inode);
3410 if (ret && ret != -ESTALE)
3411 goto out;
3412
3413 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3414 struct btrfs_root *dead_root;
3415 struct btrfs_fs_info *fs_info = root->fs_info;
3416 int is_dead_root = 0;
3417
3418 /*
3419 * this is an orphan in the tree root. Currently these
3420 * could come from 2 sources:
3421 * a) a snapshot deletion in progress
3422 * b) a free space cache inode
3423 * We need to distinguish those two, as the snapshot
3424 * orphan must not get deleted.
3425 * find_dead_roots already ran before us, so if this
3426 * is a snapshot deletion, we should find the root
3427 * in the dead_roots list
3428 */
3429 spin_lock(&fs_info->trans_lock);
3430 list_for_each_entry(dead_root, &fs_info->dead_roots,
3431 root_list) {
3432 if (dead_root->root_key.objectid ==
3433 found_key.objectid) {
3434 is_dead_root = 1;
3435 break;
3436 }
3437 }
3438 spin_unlock(&fs_info->trans_lock);
3439 if (is_dead_root) {
3440 /* prevent this orphan from being found again */
3441 key.offset = found_key.objectid - 1;
3442 continue;
3443 }
3444 }
3445 /*
3446 * Inode is already gone but the orphan item is still there,
3447 * kill the orphan item.
3448 */
3449 if (ret == -ESTALE) {
3450 trans = btrfs_start_transaction(root, 1);
3451 if (IS_ERR(trans)) {
3452 ret = PTR_ERR(trans);
3453 goto out;
3454 }
3455 btrfs_debug(root->fs_info, "auto deleting %Lu",
3456 found_key.objectid);
3457 ret = btrfs_del_orphan_item(trans, root,
3458 found_key.objectid);
3459 btrfs_end_transaction(trans, root);
3460 if (ret)
3461 goto out;
3462 continue;
3463 }
3464
3465 /*
3466 * add this inode to the orphan list so btrfs_orphan_del does
3467 * the proper thing when we hit it
3468 */
3469 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3470 &BTRFS_I(inode)->runtime_flags);
3471 atomic_inc(&root->orphan_inodes);
3472
3473 /* if we have links, this was a truncate, lets do that */
3474 if (inode->i_nlink) {
3475 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3476 iput(inode);
3477 continue;
3478 }
3479 nr_truncate++;
3480
3481 /* 1 for the orphan item deletion. */
3482 trans = btrfs_start_transaction(root, 1);
3483 if (IS_ERR(trans)) {
3484 iput(inode);
3485 ret = PTR_ERR(trans);
3486 goto out;
3487 }
3488 ret = btrfs_orphan_add(trans, inode);
3489 btrfs_end_transaction(trans, root);
3490 if (ret) {
3491 iput(inode);
3492 goto out;
3493 }
3494
3495 ret = btrfs_truncate(inode);
3496 if (ret)
3497 btrfs_orphan_del(NULL, inode);
3498 } else {
3499 nr_unlink++;
3500 }
3501
3502 /* this will do delete_inode and everything for us */
3503 iput(inode);
3504 if (ret)
3505 goto out;
3506 }
3507 /* release the path since we're done with it */
3508 btrfs_release_path(path);
3509
3510 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3511
3512 if (root->orphan_block_rsv)
3513 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3514 (u64)-1);
3515
3516 if (root->orphan_block_rsv ||
3517 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3518 trans = btrfs_join_transaction(root);
3519 if (!IS_ERR(trans))
3520 btrfs_end_transaction(trans, root);
3521 }
3522
3523 if (nr_unlink)
3524 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3525 if (nr_truncate)
3526 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3527
3528 out:
3529 if (ret)
3530 btrfs_err(root->fs_info,
3531 "could not do orphan cleanup %d", ret);
3532 btrfs_free_path(path);
3533 return ret;
3534 }
3535
3536 /*
3537 * very simple check to peek ahead in the leaf looking for xattrs. If we
3538 * don't find any xattrs, we know there can't be any acls.
3539 *
3540 * slot is the slot the inode is in, objectid is the objectid of the inode
3541 */
3542 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3543 int slot, u64 objectid,
3544 int *first_xattr_slot)
3545 {
3546 u32 nritems = btrfs_header_nritems(leaf);
3547 struct btrfs_key found_key;
3548 static u64 xattr_access = 0;
3549 static u64 xattr_default = 0;
3550 int scanned = 0;
3551
3552 if (!xattr_access) {
3553 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3554 strlen(POSIX_ACL_XATTR_ACCESS));
3555 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3556 strlen(POSIX_ACL_XATTR_DEFAULT));
3557 }
3558
3559 slot++;
3560 *first_xattr_slot = -1;
3561 while (slot < nritems) {
3562 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3563
3564 /* we found a different objectid, there must not be acls */
3565 if (found_key.objectid != objectid)
3566 return 0;
3567
3568 /* we found an xattr, assume we've got an acl */
3569 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3570 if (*first_xattr_slot == -1)
3571 *first_xattr_slot = slot;
3572 if (found_key.offset == xattr_access ||
3573 found_key.offset == xattr_default)
3574 return 1;
3575 }
3576
3577 /*
3578 * we found a key greater than an xattr key, there can't
3579 * be any acls later on
3580 */
3581 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3582 return 0;
3583
3584 slot++;
3585 scanned++;
3586
3587 /*
3588 * it goes inode, inode backrefs, xattrs, extents,
3589 * so if there are a ton of hard links to an inode there can
3590 * be a lot of backrefs. Don't waste time searching too hard,
3591 * this is just an optimization
3592 */
3593 if (scanned >= 8)
3594 break;
3595 }
3596 /* we hit the end of the leaf before we found an xattr or
3597 * something larger than an xattr. We have to assume the inode
3598 * has acls
3599 */
3600 if (*first_xattr_slot == -1)
3601 *first_xattr_slot = slot;
3602 return 1;
3603 }
3604
3605 /*
3606 * read an inode from the btree into the in-memory inode
3607 */
3608 static void btrfs_read_locked_inode(struct inode *inode)
3609 {
3610 struct btrfs_path *path;
3611 struct extent_buffer *leaf;
3612 struct btrfs_inode_item *inode_item;
3613 struct btrfs_root *root = BTRFS_I(inode)->root;
3614 struct btrfs_key location;
3615 unsigned long ptr;
3616 int maybe_acls;
3617 u32 rdev;
3618 int ret;
3619 bool filled = false;
3620 int first_xattr_slot;
3621
3622 ret = btrfs_fill_inode(inode, &rdev);
3623 if (!ret)
3624 filled = true;
3625
3626 path = btrfs_alloc_path();
3627 if (!path)
3628 goto make_bad;
3629
3630 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3631
3632 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3633 if (ret)
3634 goto make_bad;
3635
3636 leaf = path->nodes[0];
3637
3638 if (filled)
3639 goto cache_index;
3640
3641 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3642 struct btrfs_inode_item);
3643 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3644 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3645 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3646 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3647 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3648
3649 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3650 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3651
3652 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3653 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3654
3655 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3656 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3657
3658 BTRFS_I(inode)->i_otime.tv_sec =
3659 btrfs_timespec_sec(leaf, &inode_item->otime);
3660 BTRFS_I(inode)->i_otime.tv_nsec =
3661 btrfs_timespec_nsec(leaf, &inode_item->otime);
3662
3663 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3664 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3665 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3666
3667 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3668 inode->i_generation = BTRFS_I(inode)->generation;
3669 inode->i_rdev = 0;
3670 rdev = btrfs_inode_rdev(leaf, inode_item);
3671
3672 BTRFS_I(inode)->index_cnt = (u64)-1;
3673 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3674
3675 cache_index:
3676 /*
3677 * If we were modified in the current generation and evicted from memory
3678 * and then re-read we need to do a full sync since we don't have any
3679 * idea about which extents were modified before we were evicted from
3680 * cache.
3681 *
3682 * This is required for both inode re-read from disk and delayed inode
3683 * in delayed_nodes_tree.
3684 */
3685 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3686 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3687 &BTRFS_I(inode)->runtime_flags);
3688
3689 /*
3690 * We don't persist the id of the transaction where an unlink operation
3691 * against the inode was last made. So here we assume the inode might
3692 * have been evicted, and therefore the exact value of last_unlink_trans
3693 * lost, and set it to last_trans to avoid metadata inconsistencies
3694 * between the inode and its parent if the inode is fsync'ed and the log
3695 * replayed. For example, in the scenario:
3696 *
3697 * touch mydir/foo
3698 * ln mydir/foo mydir/bar
3699 * sync
3700 * unlink mydir/bar
3701 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3702 * xfs_io -c fsync mydir/foo
3703 * <power failure>
3704 * mount fs, triggers fsync log replay
3705 *
3706 * We must make sure that when we fsync our inode foo we also log its
3707 * parent inode, otherwise after log replay the parent still has the
3708 * dentry with the "bar" name but our inode foo has a link count of 1
3709 * and doesn't have an inode ref with the name "bar" anymore.
3710 *
3711 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3712 * but it guarantees correctness at the expense of ocassional full
3713 * transaction commits on fsync if our inode is a directory, or if our
3714 * inode is not a directory, logging its parent unnecessarily.
3715 */
3716 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3717
3718 path->slots[0]++;
3719 if (inode->i_nlink != 1 ||
3720 path->slots[0] >= btrfs_header_nritems(leaf))
3721 goto cache_acl;
3722
3723 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3724 if (location.objectid != btrfs_ino(inode))
3725 goto cache_acl;
3726
3727 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3728 if (location.type == BTRFS_INODE_REF_KEY) {
3729 struct btrfs_inode_ref *ref;
3730
3731 ref = (struct btrfs_inode_ref *)ptr;
3732 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3733 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3734 struct btrfs_inode_extref *extref;
3735
3736 extref = (struct btrfs_inode_extref *)ptr;
3737 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3738 extref);
3739 }
3740 cache_acl:
3741 /*
3742 * try to precache a NULL acl entry for files that don't have
3743 * any xattrs or acls
3744 */
3745 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3746 btrfs_ino(inode), &first_xattr_slot);
3747 if (first_xattr_slot != -1) {
3748 path->slots[0] = first_xattr_slot;
3749 ret = btrfs_load_inode_props(inode, path);
3750 if (ret)
3751 btrfs_err(root->fs_info,
3752 "error loading props for ino %llu (root %llu): %d",
3753 btrfs_ino(inode),
3754 root->root_key.objectid, ret);
3755 }
3756 btrfs_free_path(path);
3757
3758 if (!maybe_acls)
3759 cache_no_acl(inode);
3760
3761 switch (inode->i_mode & S_IFMT) {
3762 case S_IFREG:
3763 inode->i_mapping->a_ops = &btrfs_aops;
3764 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3765 inode->i_fop = &btrfs_file_operations;
3766 inode->i_op = &btrfs_file_inode_operations;
3767 break;
3768 case S_IFDIR:
3769 inode->i_fop = &btrfs_dir_file_operations;
3770 if (root == root->fs_info->tree_root)
3771 inode->i_op = &btrfs_dir_ro_inode_operations;
3772 else
3773 inode->i_op = &btrfs_dir_inode_operations;
3774 break;
3775 case S_IFLNK:
3776 inode->i_op = &btrfs_symlink_inode_operations;
3777 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3778 break;
3779 default:
3780 inode->i_op = &btrfs_special_inode_operations;
3781 init_special_inode(inode, inode->i_mode, rdev);
3782 break;
3783 }
3784
3785 btrfs_update_iflags(inode);
3786 return;
3787
3788 make_bad:
3789 btrfs_free_path(path);
3790 make_bad_inode(inode);
3791 }
3792
3793 /*
3794 * given a leaf and an inode, copy the inode fields into the leaf
3795 */
3796 static void fill_inode_item(struct btrfs_trans_handle *trans,
3797 struct extent_buffer *leaf,
3798 struct btrfs_inode_item *item,
3799 struct inode *inode)
3800 {
3801 struct btrfs_map_token token;
3802
3803 btrfs_init_map_token(&token);
3804
3805 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3806 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3807 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3808 &token);
3809 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3810 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3811
3812 btrfs_set_token_timespec_sec(leaf, &item->atime,
3813 inode->i_atime.tv_sec, &token);
3814 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3815 inode->i_atime.tv_nsec, &token);
3816
3817 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3818 inode->i_mtime.tv_sec, &token);
3819 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3820 inode->i_mtime.tv_nsec, &token);
3821
3822 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3823 inode->i_ctime.tv_sec, &token);
3824 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3825 inode->i_ctime.tv_nsec, &token);
3826
3827 btrfs_set_token_timespec_sec(leaf, &item->otime,
3828 BTRFS_I(inode)->i_otime.tv_sec, &token);
3829 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3830 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3831
3832 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3833 &token);
3834 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3835 &token);
3836 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3837 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3838 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3839 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3840 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3841 }
3842
3843 /*
3844 * copy everything in the in-memory inode into the btree.
3845 */
3846 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3847 struct btrfs_root *root, struct inode *inode)
3848 {
3849 struct btrfs_inode_item *inode_item;
3850 struct btrfs_path *path;
3851 struct extent_buffer *leaf;
3852 int ret;
3853
3854 path = btrfs_alloc_path();
3855 if (!path)
3856 return -ENOMEM;
3857
3858 path->leave_spinning = 1;
3859 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3860 1);
3861 if (ret) {
3862 if (ret > 0)
3863 ret = -ENOENT;
3864 goto failed;
3865 }
3866
3867 leaf = path->nodes[0];
3868 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3869 struct btrfs_inode_item);
3870
3871 fill_inode_item(trans, leaf, inode_item, inode);
3872 btrfs_mark_buffer_dirty(leaf);
3873 btrfs_set_inode_last_trans(trans, inode);
3874 ret = 0;
3875 failed:
3876 btrfs_free_path(path);
3877 return ret;
3878 }
3879
3880 /*
3881 * copy everything in the in-memory inode into the btree.
3882 */
3883 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3884 struct btrfs_root *root, struct inode *inode)
3885 {
3886 int ret;
3887
3888 /*
3889 * If the inode is a free space inode, we can deadlock during commit
3890 * if we put it into the delayed code.
3891 *
3892 * The data relocation inode should also be directly updated
3893 * without delay
3894 */
3895 if (!btrfs_is_free_space_inode(inode)
3896 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3897 && !root->fs_info->log_root_recovering) {
3898 btrfs_update_root_times(trans, root);
3899
3900 ret = btrfs_delayed_update_inode(trans, root, inode);
3901 if (!ret)
3902 btrfs_set_inode_last_trans(trans, inode);
3903 return ret;
3904 }
3905
3906 return btrfs_update_inode_item(trans, root, inode);
3907 }
3908
3909 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3910 struct btrfs_root *root,
3911 struct inode *inode)
3912 {
3913 int ret;
3914
3915 ret = btrfs_update_inode(trans, root, inode);
3916 if (ret == -ENOSPC)
3917 return btrfs_update_inode_item(trans, root, inode);
3918 return ret;
3919 }
3920
3921 /*
3922 * unlink helper that gets used here in inode.c and in the tree logging
3923 * recovery code. It remove a link in a directory with a given name, and
3924 * also drops the back refs in the inode to the directory
3925 */
3926 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3927 struct btrfs_root *root,
3928 struct inode *dir, struct inode *inode,
3929 const char *name, int name_len)
3930 {
3931 struct btrfs_path *path;
3932 int ret = 0;
3933 struct extent_buffer *leaf;
3934 struct btrfs_dir_item *di;
3935 struct btrfs_key key;
3936 u64 index;
3937 u64 ino = btrfs_ino(inode);
3938 u64 dir_ino = btrfs_ino(dir);
3939
3940 path = btrfs_alloc_path();
3941 if (!path) {
3942 ret = -ENOMEM;
3943 goto out;
3944 }
3945
3946 path->leave_spinning = 1;
3947 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3948 name, name_len, -1);
3949 if (IS_ERR(di)) {
3950 ret = PTR_ERR(di);
3951 goto err;
3952 }
3953 if (!di) {
3954 ret = -ENOENT;
3955 goto err;
3956 }
3957 leaf = path->nodes[0];
3958 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3959 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3960 if (ret)
3961 goto err;
3962 btrfs_release_path(path);
3963
3964 /*
3965 * If we don't have dir index, we have to get it by looking up
3966 * the inode ref, since we get the inode ref, remove it directly,
3967 * it is unnecessary to do delayed deletion.
3968 *
3969 * But if we have dir index, needn't search inode ref to get it.
3970 * Since the inode ref is close to the inode item, it is better
3971 * that we delay to delete it, and just do this deletion when
3972 * we update the inode item.
3973 */
3974 if (BTRFS_I(inode)->dir_index) {
3975 ret = btrfs_delayed_delete_inode_ref(inode);
3976 if (!ret) {
3977 index = BTRFS_I(inode)->dir_index;
3978 goto skip_backref;
3979 }
3980 }
3981
3982 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3983 dir_ino, &index);
3984 if (ret) {
3985 btrfs_info(root->fs_info,
3986 "failed to delete reference to %.*s, inode %llu parent %llu",
3987 name_len, name, ino, dir_ino);
3988 btrfs_abort_transaction(trans, root, ret);
3989 goto err;
3990 }
3991 skip_backref:
3992 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3993 if (ret) {
3994 btrfs_abort_transaction(trans, root, ret);
3995 goto err;
3996 }
3997
3998 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3999 inode, dir_ino);
4000 if (ret != 0 && ret != -ENOENT) {
4001 btrfs_abort_transaction(trans, root, ret);
4002 goto err;
4003 }
4004
4005 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4006 dir, index);
4007 if (ret == -ENOENT)
4008 ret = 0;
4009 else if (ret)
4010 btrfs_abort_transaction(trans, root, ret);
4011 err:
4012 btrfs_free_path(path);
4013 if (ret)
4014 goto out;
4015
4016 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4017 inode_inc_iversion(inode);
4018 inode_inc_iversion(dir);
4019 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4020 ret = btrfs_update_inode(trans, root, dir);
4021 out:
4022 return ret;
4023 }
4024
4025 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4026 struct btrfs_root *root,
4027 struct inode *dir, struct inode *inode,
4028 const char *name, int name_len)
4029 {
4030 int ret;
4031 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4032 if (!ret) {
4033 drop_nlink(inode);
4034 ret = btrfs_update_inode(trans, root, inode);
4035 }
4036 return ret;
4037 }
4038
4039 /*
4040 * helper to start transaction for unlink and rmdir.
4041 *
4042 * unlink and rmdir are special in btrfs, they do not always free space, so
4043 * if we cannot make our reservations the normal way try and see if there is
4044 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4045 * allow the unlink to occur.
4046 */
4047 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4048 {
4049 struct btrfs_root *root = BTRFS_I(dir)->root;
4050
4051 /*
4052 * 1 for the possible orphan item
4053 * 1 for the dir item
4054 * 1 for the dir index
4055 * 1 for the inode ref
4056 * 1 for the inode
4057 */
4058 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4059 }
4060
4061 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4062 {
4063 struct btrfs_root *root = BTRFS_I(dir)->root;
4064 struct btrfs_trans_handle *trans;
4065 struct inode *inode = d_inode(dentry);
4066 int ret;
4067
4068 trans = __unlink_start_trans(dir);
4069 if (IS_ERR(trans))
4070 return PTR_ERR(trans);
4071
4072 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4073
4074 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4075 dentry->d_name.name, dentry->d_name.len);
4076 if (ret)
4077 goto out;
4078
4079 if (inode->i_nlink == 0) {
4080 ret = btrfs_orphan_add(trans, inode);
4081 if (ret)
4082 goto out;
4083 }
4084
4085 out:
4086 btrfs_end_transaction(trans, root);
4087 btrfs_btree_balance_dirty(root);
4088 return ret;
4089 }
4090
4091 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4092 struct btrfs_root *root,
4093 struct inode *dir, u64 objectid,
4094 const char *name, int name_len)
4095 {
4096 struct btrfs_path *path;
4097 struct extent_buffer *leaf;
4098 struct btrfs_dir_item *di;
4099 struct btrfs_key key;
4100 u64 index;
4101 int ret;
4102 u64 dir_ino = btrfs_ino(dir);
4103
4104 path = btrfs_alloc_path();
4105 if (!path)
4106 return -ENOMEM;
4107
4108 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4109 name, name_len, -1);
4110 if (IS_ERR_OR_NULL(di)) {
4111 if (!di)
4112 ret = -ENOENT;
4113 else
4114 ret = PTR_ERR(di);
4115 goto out;
4116 }
4117
4118 leaf = path->nodes[0];
4119 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4120 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4121 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4122 if (ret) {
4123 btrfs_abort_transaction(trans, root, ret);
4124 goto out;
4125 }
4126 btrfs_release_path(path);
4127
4128 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4129 objectid, root->root_key.objectid,
4130 dir_ino, &index, name, name_len);
4131 if (ret < 0) {
4132 if (ret != -ENOENT) {
4133 btrfs_abort_transaction(trans, root, ret);
4134 goto out;
4135 }
4136 di = btrfs_search_dir_index_item(root, path, dir_ino,
4137 name, name_len);
4138 if (IS_ERR_OR_NULL(di)) {
4139 if (!di)
4140 ret = -ENOENT;
4141 else
4142 ret = PTR_ERR(di);
4143 btrfs_abort_transaction(trans, root, ret);
4144 goto out;
4145 }
4146
4147 leaf = path->nodes[0];
4148 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4149 btrfs_release_path(path);
4150 index = key.offset;
4151 }
4152 btrfs_release_path(path);
4153
4154 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4155 if (ret) {
4156 btrfs_abort_transaction(trans, root, ret);
4157 goto out;
4158 }
4159
4160 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4161 inode_inc_iversion(dir);
4162 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4163 ret = btrfs_update_inode_fallback(trans, root, dir);
4164 if (ret)
4165 btrfs_abort_transaction(trans, root, ret);
4166 out:
4167 btrfs_free_path(path);
4168 return ret;
4169 }
4170
4171 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4172 {
4173 struct inode *inode = d_inode(dentry);
4174 int err = 0;
4175 struct btrfs_root *root = BTRFS_I(dir)->root;
4176 struct btrfs_trans_handle *trans;
4177
4178 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4179 return -ENOTEMPTY;
4180 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4181 return -EPERM;
4182
4183 trans = __unlink_start_trans(dir);
4184 if (IS_ERR(trans))
4185 return PTR_ERR(trans);
4186
4187 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4188 err = btrfs_unlink_subvol(trans, root, dir,
4189 BTRFS_I(inode)->location.objectid,
4190 dentry->d_name.name,
4191 dentry->d_name.len);
4192 goto out;
4193 }
4194
4195 err = btrfs_orphan_add(trans, inode);
4196 if (err)
4197 goto out;
4198
4199 /* now the directory is empty */
4200 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4201 dentry->d_name.name, dentry->d_name.len);
4202 if (!err)
4203 btrfs_i_size_write(inode, 0);
4204 out:
4205 btrfs_end_transaction(trans, root);
4206 btrfs_btree_balance_dirty(root);
4207
4208 return err;
4209 }
4210
4211 static int truncate_space_check(struct btrfs_trans_handle *trans,
4212 struct btrfs_root *root,
4213 u64 bytes_deleted)
4214 {
4215 int ret;
4216
4217 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4218 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4219 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4220 if (!ret)
4221 trans->bytes_reserved += bytes_deleted;
4222 return ret;
4223
4224 }
4225
4226 static int truncate_inline_extent(struct inode *inode,
4227 struct btrfs_path *path,
4228 struct btrfs_key *found_key,
4229 const u64 item_end,
4230 const u64 new_size)
4231 {
4232 struct extent_buffer *leaf = path->nodes[0];
4233 int slot = path->slots[0];
4234 struct btrfs_file_extent_item *fi;
4235 u32 size = (u32)(new_size - found_key->offset);
4236 struct btrfs_root *root = BTRFS_I(inode)->root;
4237
4238 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4239
4240 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4241 loff_t offset = new_size;
4242 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4243
4244 /*
4245 * Zero out the remaining of the last page of our inline extent,
4246 * instead of directly truncating our inline extent here - that
4247 * would be much more complex (decompressing all the data, then
4248 * compressing the truncated data, which might be bigger than
4249 * the size of the inline extent, resize the extent, etc).
4250 * We release the path because to get the page we might need to
4251 * read the extent item from disk (data not in the page cache).
4252 */
4253 btrfs_release_path(path);
4254 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4255 }
4256
4257 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4258 size = btrfs_file_extent_calc_inline_size(size);
4259 btrfs_truncate_item(root, path, size, 1);
4260
4261 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4262 inode_sub_bytes(inode, item_end + 1 - new_size);
4263
4264 return 0;
4265 }
4266
4267 /*
4268 * this can truncate away extent items, csum items and directory items.
4269 * It starts at a high offset and removes keys until it can't find
4270 * any higher than new_size
4271 *
4272 * csum items that cross the new i_size are truncated to the new size
4273 * as well.
4274 *
4275 * min_type is the minimum key type to truncate down to. If set to 0, this
4276 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4277 */
4278 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4279 struct btrfs_root *root,
4280 struct inode *inode,
4281 u64 new_size, u32 min_type)
4282 {
4283 struct btrfs_path *path;
4284 struct extent_buffer *leaf;
4285 struct btrfs_file_extent_item *fi;
4286 struct btrfs_key key;
4287 struct btrfs_key found_key;
4288 u64 extent_start = 0;
4289 u64 extent_num_bytes = 0;
4290 u64 extent_offset = 0;
4291 u64 item_end = 0;
4292 u64 last_size = new_size;
4293 u32 found_type = (u8)-1;
4294 int found_extent;
4295 int del_item;
4296 int pending_del_nr = 0;
4297 int pending_del_slot = 0;
4298 int extent_type = -1;
4299 int ret;
4300 int err = 0;
4301 u64 ino = btrfs_ino(inode);
4302 u64 bytes_deleted = 0;
4303 bool be_nice = 0;
4304 bool should_throttle = 0;
4305 bool should_end = 0;
4306
4307 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4308
4309 /*
4310 * for non-free space inodes and ref cows, we want to back off from
4311 * time to time
4312 */
4313 if (!btrfs_is_free_space_inode(inode) &&
4314 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4315 be_nice = 1;
4316
4317 path = btrfs_alloc_path();
4318 if (!path)
4319 return -ENOMEM;
4320 path->reada = -1;
4321
4322 /*
4323 * We want to drop from the next block forward in case this new size is
4324 * not block aligned since we will be keeping the last block of the
4325 * extent just the way it is.
4326 */
4327 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4328 root == root->fs_info->tree_root)
4329 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4330 root->sectorsize), (u64)-1, 0);
4331
4332 /*
4333 * This function is also used to drop the items in the log tree before
4334 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4335 * it is used to drop the loged items. So we shouldn't kill the delayed
4336 * items.
4337 */
4338 if (min_type == 0 && root == BTRFS_I(inode)->root)
4339 btrfs_kill_delayed_inode_items(inode);
4340
4341 key.objectid = ino;
4342 key.offset = (u64)-1;
4343 key.type = (u8)-1;
4344
4345 search_again:
4346 /*
4347 * with a 16K leaf size and 128MB extents, you can actually queue
4348 * up a huge file in a single leaf. Most of the time that
4349 * bytes_deleted is > 0, it will be huge by the time we get here
4350 */
4351 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4352 if (btrfs_should_end_transaction(trans, root)) {
4353 err = -EAGAIN;
4354 goto error;
4355 }
4356 }
4357
4358
4359 path->leave_spinning = 1;
4360 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4361 if (ret < 0) {
4362 err = ret;
4363 goto out;
4364 }
4365
4366 if (ret > 0) {
4367 /* there are no items in the tree for us to truncate, we're
4368 * done
4369 */
4370 if (path->slots[0] == 0)
4371 goto out;
4372 path->slots[0]--;
4373 }
4374
4375 while (1) {
4376 fi = NULL;
4377 leaf = path->nodes[0];
4378 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4379 found_type = found_key.type;
4380
4381 if (found_key.objectid != ino)
4382 break;
4383
4384 if (found_type < min_type)
4385 break;
4386
4387 item_end = found_key.offset;
4388 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4389 fi = btrfs_item_ptr(leaf, path->slots[0],
4390 struct btrfs_file_extent_item);
4391 extent_type = btrfs_file_extent_type(leaf, fi);
4392 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4393 item_end +=
4394 btrfs_file_extent_num_bytes(leaf, fi);
4395 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4396 item_end += btrfs_file_extent_inline_len(leaf,
4397 path->slots[0], fi);
4398 }
4399 item_end--;
4400 }
4401 if (found_type > min_type) {
4402 del_item = 1;
4403 } else {
4404 if (item_end < new_size)
4405 break;
4406 if (found_key.offset >= new_size)
4407 del_item = 1;
4408 else
4409 del_item = 0;
4410 }
4411 found_extent = 0;
4412 /* FIXME, shrink the extent if the ref count is only 1 */
4413 if (found_type != BTRFS_EXTENT_DATA_KEY)
4414 goto delete;
4415
4416 if (del_item)
4417 last_size = found_key.offset;
4418 else
4419 last_size = new_size;
4420
4421 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4422 u64 num_dec;
4423 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4424 if (!del_item) {
4425 u64 orig_num_bytes =
4426 btrfs_file_extent_num_bytes(leaf, fi);
4427 extent_num_bytes = ALIGN(new_size -
4428 found_key.offset,
4429 root->sectorsize);
4430 btrfs_set_file_extent_num_bytes(leaf, fi,
4431 extent_num_bytes);
4432 num_dec = (orig_num_bytes -
4433 extent_num_bytes);
4434 if (test_bit(BTRFS_ROOT_REF_COWS,
4435 &root->state) &&
4436 extent_start != 0)
4437 inode_sub_bytes(inode, num_dec);
4438 btrfs_mark_buffer_dirty(leaf);
4439 } else {
4440 extent_num_bytes =
4441 btrfs_file_extent_disk_num_bytes(leaf,
4442 fi);
4443 extent_offset = found_key.offset -
4444 btrfs_file_extent_offset(leaf, fi);
4445
4446 /* FIXME blocksize != 4096 */
4447 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4448 if (extent_start != 0) {
4449 found_extent = 1;
4450 if (test_bit(BTRFS_ROOT_REF_COWS,
4451 &root->state))
4452 inode_sub_bytes(inode, num_dec);
4453 }
4454 }
4455 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4456 /*
4457 * we can't truncate inline items that have had
4458 * special encodings
4459 */
4460 if (!del_item &&
4461 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4462 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4463
4464 /*
4465 * Need to release path in order to truncate a
4466 * compressed extent. So delete any accumulated
4467 * extent items so far.
4468 */
4469 if (btrfs_file_extent_compression(leaf, fi) !=
4470 BTRFS_COMPRESS_NONE && pending_del_nr) {
4471 err = btrfs_del_items(trans, root, path,
4472 pending_del_slot,
4473 pending_del_nr);
4474 if (err) {
4475 btrfs_abort_transaction(trans,
4476 root,
4477 err);
4478 goto error;
4479 }
4480 pending_del_nr = 0;
4481 }
4482
4483 err = truncate_inline_extent(inode, path,
4484 &found_key,
4485 item_end,
4486 new_size);
4487 if (err) {
4488 btrfs_abort_transaction(trans,
4489 root, err);
4490 goto error;
4491 }
4492 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4493 &root->state)) {
4494 inode_sub_bytes(inode, item_end + 1 - new_size);
4495 }
4496 }
4497 delete:
4498 if (del_item) {
4499 if (!pending_del_nr) {
4500 /* no pending yet, add ourselves */
4501 pending_del_slot = path->slots[0];
4502 pending_del_nr = 1;
4503 } else if (pending_del_nr &&
4504 path->slots[0] + 1 == pending_del_slot) {
4505 /* hop on the pending chunk */
4506 pending_del_nr++;
4507 pending_del_slot = path->slots[0];
4508 } else {
4509 BUG();
4510 }
4511 } else {
4512 break;
4513 }
4514 should_throttle = 0;
4515
4516 if (found_extent &&
4517 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4518 root == root->fs_info->tree_root)) {
4519 btrfs_set_path_blocking(path);
4520 bytes_deleted += extent_num_bytes;
4521 ret = btrfs_free_extent(trans, root, extent_start,
4522 extent_num_bytes, 0,
4523 btrfs_header_owner(leaf),
4524 ino, extent_offset);
4525 BUG_ON(ret);
4526 if (btrfs_should_throttle_delayed_refs(trans, root))
4527 btrfs_async_run_delayed_refs(root,
4528 trans->delayed_ref_updates * 2, 0);
4529 if (be_nice) {
4530 if (truncate_space_check(trans, root,
4531 extent_num_bytes)) {
4532 should_end = 1;
4533 }
4534 if (btrfs_should_throttle_delayed_refs(trans,
4535 root)) {
4536 should_throttle = 1;
4537 }
4538 }
4539 }
4540
4541 if (found_type == BTRFS_INODE_ITEM_KEY)
4542 break;
4543
4544 if (path->slots[0] == 0 ||
4545 path->slots[0] != pending_del_slot ||
4546 should_throttle || should_end) {
4547 if (pending_del_nr) {
4548 ret = btrfs_del_items(trans, root, path,
4549 pending_del_slot,
4550 pending_del_nr);
4551 if (ret) {
4552 btrfs_abort_transaction(trans,
4553 root, ret);
4554 goto error;
4555 }
4556 pending_del_nr = 0;
4557 }
4558 btrfs_release_path(path);
4559 if (should_throttle) {
4560 unsigned long updates = trans->delayed_ref_updates;
4561 if (updates) {
4562 trans->delayed_ref_updates = 0;
4563 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4564 if (ret && !err)
4565 err = ret;
4566 }
4567 }
4568 /*
4569 * if we failed to refill our space rsv, bail out
4570 * and let the transaction restart
4571 */
4572 if (should_end) {
4573 err = -EAGAIN;
4574 goto error;
4575 }
4576 goto search_again;
4577 } else {
4578 path->slots[0]--;
4579 }
4580 }
4581 out:
4582 if (pending_del_nr) {
4583 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4584 pending_del_nr);
4585 if (ret)
4586 btrfs_abort_transaction(trans, root, ret);
4587 }
4588 error:
4589 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4590 btrfs_ordered_update_i_size(inode, last_size, NULL);
4591
4592 btrfs_free_path(path);
4593
4594 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4595 unsigned long updates = trans->delayed_ref_updates;
4596 if (updates) {
4597 trans->delayed_ref_updates = 0;
4598 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4599 if (ret && !err)
4600 err = ret;
4601 }
4602 }
4603 return err;
4604 }
4605
4606 /*
4607 * btrfs_truncate_page - read, zero a chunk and write a page
4608 * @inode - inode that we're zeroing
4609 * @from - the offset to start zeroing
4610 * @len - the length to zero, 0 to zero the entire range respective to the
4611 * offset
4612 * @front - zero up to the offset instead of from the offset on
4613 *
4614 * This will find the page for the "from" offset and cow the page and zero the
4615 * part we want to zero. This is used with truncate and hole punching.
4616 */
4617 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4618 int front)
4619 {
4620 struct address_space *mapping = inode->i_mapping;
4621 struct btrfs_root *root = BTRFS_I(inode)->root;
4622 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4623 struct btrfs_ordered_extent *ordered;
4624 struct extent_state *cached_state = NULL;
4625 char *kaddr;
4626 u32 blocksize = root->sectorsize;
4627 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4628 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4629 struct page *page;
4630 gfp_t mask = btrfs_alloc_write_mask(mapping);
4631 int ret = 0;
4632 u64 page_start;
4633 u64 page_end;
4634
4635 if ((offset & (blocksize - 1)) == 0 &&
4636 (!len || ((len & (blocksize - 1)) == 0)))
4637 goto out;
4638 ret = btrfs_delalloc_reserve_space(inode,
4639 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4640 if (ret)
4641 goto out;
4642
4643 again:
4644 page = find_or_create_page(mapping, index, mask);
4645 if (!page) {
4646 btrfs_delalloc_release_space(inode,
4647 round_down(from, PAGE_CACHE_SIZE),
4648 PAGE_CACHE_SIZE);
4649 ret = -ENOMEM;
4650 goto out;
4651 }
4652
4653 page_start = page_offset(page);
4654 page_end = page_start + PAGE_CACHE_SIZE - 1;
4655
4656 if (!PageUptodate(page)) {
4657 ret = btrfs_readpage(NULL, page);
4658 lock_page(page);
4659 if (page->mapping != mapping) {
4660 unlock_page(page);
4661 page_cache_release(page);
4662 goto again;
4663 }
4664 if (!PageUptodate(page)) {
4665 ret = -EIO;
4666 goto out_unlock;
4667 }
4668 }
4669 wait_on_page_writeback(page);
4670
4671 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4672 set_page_extent_mapped(page);
4673
4674 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4675 if (ordered) {
4676 unlock_extent_cached(io_tree, page_start, page_end,
4677 &cached_state, GFP_NOFS);
4678 unlock_page(page);
4679 page_cache_release(page);
4680 btrfs_start_ordered_extent(inode, ordered, 1);
4681 btrfs_put_ordered_extent(ordered);
4682 goto again;
4683 }
4684
4685 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4686 EXTENT_DIRTY | EXTENT_DELALLOC |
4687 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4688 0, 0, &cached_state, GFP_NOFS);
4689
4690 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4691 &cached_state);
4692 if (ret) {
4693 unlock_extent_cached(io_tree, page_start, page_end,
4694 &cached_state, GFP_NOFS);
4695 goto out_unlock;
4696 }
4697
4698 if (offset != PAGE_CACHE_SIZE) {
4699 if (!len)
4700 len = PAGE_CACHE_SIZE - offset;
4701 kaddr = kmap(page);
4702 if (front)
4703 memset(kaddr, 0, offset);
4704 else
4705 memset(kaddr + offset, 0, len);
4706 flush_dcache_page(page);
4707 kunmap(page);
4708 }
4709 ClearPageChecked(page);
4710 set_page_dirty(page);
4711 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4712 GFP_NOFS);
4713
4714 out_unlock:
4715 if (ret)
4716 btrfs_delalloc_release_space(inode, page_start,
4717 PAGE_CACHE_SIZE);
4718 unlock_page(page);
4719 page_cache_release(page);
4720 out:
4721 return ret;
4722 }
4723
4724 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4725 u64 offset, u64 len)
4726 {
4727 struct btrfs_trans_handle *trans;
4728 int ret;
4729
4730 /*
4731 * Still need to make sure the inode looks like it's been updated so
4732 * that any holes get logged if we fsync.
4733 */
4734 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4735 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4736 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4737 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4738 return 0;
4739 }
4740
4741 /*
4742 * 1 - for the one we're dropping
4743 * 1 - for the one we're adding
4744 * 1 - for updating the inode.
4745 */
4746 trans = btrfs_start_transaction(root, 3);
4747 if (IS_ERR(trans))
4748 return PTR_ERR(trans);
4749
4750 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4751 if (ret) {
4752 btrfs_abort_transaction(trans, root, ret);
4753 btrfs_end_transaction(trans, root);
4754 return ret;
4755 }
4756
4757 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4758 0, 0, len, 0, len, 0, 0, 0);
4759 if (ret)
4760 btrfs_abort_transaction(trans, root, ret);
4761 else
4762 btrfs_update_inode(trans, root, inode);
4763 btrfs_end_transaction(trans, root);
4764 return ret;
4765 }
4766
4767 /*
4768 * This function puts in dummy file extents for the area we're creating a hole
4769 * for. So if we are truncating this file to a larger size we need to insert
4770 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4771 * the range between oldsize and size
4772 */
4773 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4774 {
4775 struct btrfs_root *root = BTRFS_I(inode)->root;
4776 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4777 struct extent_map *em = NULL;
4778 struct extent_state *cached_state = NULL;
4779 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4780 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4781 u64 block_end = ALIGN(size, root->sectorsize);
4782 u64 last_byte;
4783 u64 cur_offset;
4784 u64 hole_size;
4785 int err = 0;
4786
4787 /*
4788 * If our size started in the middle of a page we need to zero out the
4789 * rest of the page before we expand the i_size, otherwise we could
4790 * expose stale data.
4791 */
4792 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4793 if (err)
4794 return err;
4795
4796 if (size <= hole_start)
4797 return 0;
4798
4799 while (1) {
4800 struct btrfs_ordered_extent *ordered;
4801
4802 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4803 &cached_state);
4804 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4805 block_end - hole_start);
4806 if (!ordered)
4807 break;
4808 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4809 &cached_state, GFP_NOFS);
4810 btrfs_start_ordered_extent(inode, ordered, 1);
4811 btrfs_put_ordered_extent(ordered);
4812 }
4813
4814 cur_offset = hole_start;
4815 while (1) {
4816 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4817 block_end - cur_offset, 0);
4818 if (IS_ERR(em)) {
4819 err = PTR_ERR(em);
4820 em = NULL;
4821 break;
4822 }
4823 last_byte = min(extent_map_end(em), block_end);
4824 last_byte = ALIGN(last_byte , root->sectorsize);
4825 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4826 struct extent_map *hole_em;
4827 hole_size = last_byte - cur_offset;
4828
4829 err = maybe_insert_hole(root, inode, cur_offset,
4830 hole_size);
4831 if (err)
4832 break;
4833 btrfs_drop_extent_cache(inode, cur_offset,
4834 cur_offset + hole_size - 1, 0);
4835 hole_em = alloc_extent_map();
4836 if (!hole_em) {
4837 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4838 &BTRFS_I(inode)->runtime_flags);
4839 goto next;
4840 }
4841 hole_em->start = cur_offset;
4842 hole_em->len = hole_size;
4843 hole_em->orig_start = cur_offset;
4844
4845 hole_em->block_start = EXTENT_MAP_HOLE;
4846 hole_em->block_len = 0;
4847 hole_em->orig_block_len = 0;
4848 hole_em->ram_bytes = hole_size;
4849 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4850 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4851 hole_em->generation = root->fs_info->generation;
4852
4853 while (1) {
4854 write_lock(&em_tree->lock);
4855 err = add_extent_mapping(em_tree, hole_em, 1);
4856 write_unlock(&em_tree->lock);
4857 if (err != -EEXIST)
4858 break;
4859 btrfs_drop_extent_cache(inode, cur_offset,
4860 cur_offset +
4861 hole_size - 1, 0);
4862 }
4863 free_extent_map(hole_em);
4864 }
4865 next:
4866 free_extent_map(em);
4867 em = NULL;
4868 cur_offset = last_byte;
4869 if (cur_offset >= block_end)
4870 break;
4871 }
4872 free_extent_map(em);
4873 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4874 GFP_NOFS);
4875 return err;
4876 }
4877
4878 static int wait_snapshoting_atomic_t(atomic_t *a)
4879 {
4880 schedule();
4881 return 0;
4882 }
4883
4884 static void wait_for_snapshot_creation(struct btrfs_root *root)
4885 {
4886 while (true) {
4887 int ret;
4888
4889 ret = btrfs_start_write_no_snapshoting(root);
4890 if (ret)
4891 break;
4892 wait_on_atomic_t(&root->will_be_snapshoted,
4893 wait_snapshoting_atomic_t,
4894 TASK_UNINTERRUPTIBLE);
4895 }
4896 }
4897
4898 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4899 {
4900 struct btrfs_root *root = BTRFS_I(inode)->root;
4901 struct btrfs_trans_handle *trans;
4902 loff_t oldsize = i_size_read(inode);
4903 loff_t newsize = attr->ia_size;
4904 int mask = attr->ia_valid;
4905 int ret;
4906
4907 /*
4908 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4909 * special case where we need to update the times despite not having
4910 * these flags set. For all other operations the VFS set these flags
4911 * explicitly if it wants a timestamp update.
4912 */
4913 if (newsize != oldsize) {
4914 inode_inc_iversion(inode);
4915 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4916 inode->i_ctime = inode->i_mtime =
4917 current_fs_time(inode->i_sb);
4918 }
4919
4920 if (newsize > oldsize) {
4921 truncate_pagecache(inode, newsize);
4922 /*
4923 * Don't do an expanding truncate while snapshoting is ongoing.
4924 * This is to ensure the snapshot captures a fully consistent
4925 * state of this file - if the snapshot captures this expanding
4926 * truncation, it must capture all writes that happened before
4927 * this truncation.
4928 */
4929 wait_for_snapshot_creation(root);
4930 ret = btrfs_cont_expand(inode, oldsize, newsize);
4931 if (ret) {
4932 btrfs_end_write_no_snapshoting(root);
4933 return ret;
4934 }
4935
4936 trans = btrfs_start_transaction(root, 1);
4937 if (IS_ERR(trans)) {
4938 btrfs_end_write_no_snapshoting(root);
4939 return PTR_ERR(trans);
4940 }
4941
4942 i_size_write(inode, newsize);
4943 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4944 ret = btrfs_update_inode(trans, root, inode);
4945 btrfs_end_write_no_snapshoting(root);
4946 btrfs_end_transaction(trans, root);
4947 } else {
4948
4949 /*
4950 * We're truncating a file that used to have good data down to
4951 * zero. Make sure it gets into the ordered flush list so that
4952 * any new writes get down to disk quickly.
4953 */
4954 if (newsize == 0)
4955 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4956 &BTRFS_I(inode)->runtime_flags);
4957
4958 /*
4959 * 1 for the orphan item we're going to add
4960 * 1 for the orphan item deletion.
4961 */
4962 trans = btrfs_start_transaction(root, 2);
4963 if (IS_ERR(trans))
4964 return PTR_ERR(trans);
4965
4966 /*
4967 * We need to do this in case we fail at _any_ point during the
4968 * actual truncate. Once we do the truncate_setsize we could
4969 * invalidate pages which forces any outstanding ordered io to
4970 * be instantly completed which will give us extents that need
4971 * to be truncated. If we fail to get an orphan inode down we
4972 * could have left over extents that were never meant to live,
4973 * so we need to garuntee from this point on that everything
4974 * will be consistent.
4975 */
4976 ret = btrfs_orphan_add(trans, inode);
4977 btrfs_end_transaction(trans, root);
4978 if (ret)
4979 return ret;
4980
4981 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4982 truncate_setsize(inode, newsize);
4983
4984 /* Disable nonlocked read DIO to avoid the end less truncate */
4985 btrfs_inode_block_unlocked_dio(inode);
4986 inode_dio_wait(inode);
4987 btrfs_inode_resume_unlocked_dio(inode);
4988
4989 ret = btrfs_truncate(inode);
4990 if (ret && inode->i_nlink) {
4991 int err;
4992
4993 /*
4994 * failed to truncate, disk_i_size is only adjusted down
4995 * as we remove extents, so it should represent the true
4996 * size of the inode, so reset the in memory size and
4997 * delete our orphan entry.
4998 */
4999 trans = btrfs_join_transaction(root);
5000 if (IS_ERR(trans)) {
5001 btrfs_orphan_del(NULL, inode);
5002 return ret;
5003 }
5004 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5005 err = btrfs_orphan_del(trans, inode);
5006 if (err)
5007 btrfs_abort_transaction(trans, root, err);
5008 btrfs_end_transaction(trans, root);
5009 }
5010 }
5011
5012 return ret;
5013 }
5014
5015 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5016 {
5017 struct inode *inode = d_inode(dentry);
5018 struct btrfs_root *root = BTRFS_I(inode)->root;
5019 int err;
5020
5021 if (btrfs_root_readonly(root))
5022 return -EROFS;
5023
5024 err = inode_change_ok(inode, attr);
5025 if (err)
5026 return err;
5027
5028 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5029 err = btrfs_setsize(inode, attr);
5030 if (err)
5031 return err;
5032 }
5033
5034 if (attr->ia_valid) {
5035 setattr_copy(inode, attr);
5036 inode_inc_iversion(inode);
5037 err = btrfs_dirty_inode(inode);
5038
5039 if (!err && attr->ia_valid & ATTR_MODE)
5040 err = posix_acl_chmod(inode, inode->i_mode);
5041 }
5042
5043 return err;
5044 }
5045
5046 /*
5047 * While truncating the inode pages during eviction, we get the VFS calling
5048 * btrfs_invalidatepage() against each page of the inode. This is slow because
5049 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5050 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5051 * extent_state structures over and over, wasting lots of time.
5052 *
5053 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5054 * those expensive operations on a per page basis and do only the ordered io
5055 * finishing, while we release here the extent_map and extent_state structures,
5056 * without the excessive merging and splitting.
5057 */
5058 static void evict_inode_truncate_pages(struct inode *inode)
5059 {
5060 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5061 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5062 struct rb_node *node;
5063
5064 ASSERT(inode->i_state & I_FREEING);
5065 truncate_inode_pages_final(&inode->i_data);
5066
5067 write_lock(&map_tree->lock);
5068 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5069 struct extent_map *em;
5070
5071 node = rb_first(&map_tree->map);
5072 em = rb_entry(node, struct extent_map, rb_node);
5073 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5074 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5075 remove_extent_mapping(map_tree, em);
5076 free_extent_map(em);
5077 if (need_resched()) {
5078 write_unlock(&map_tree->lock);
5079 cond_resched();
5080 write_lock(&map_tree->lock);
5081 }
5082 }
5083 write_unlock(&map_tree->lock);
5084
5085 /*
5086 * Keep looping until we have no more ranges in the io tree.
5087 * We can have ongoing bios started by readpages (called from readahead)
5088 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5089 * still in progress (unlocked the pages in the bio but did not yet
5090 * unlocked the ranges in the io tree). Therefore this means some
5091 * ranges can still be locked and eviction started because before
5092 * submitting those bios, which are executed by a separate task (work
5093 * queue kthread), inode references (inode->i_count) were not taken
5094 * (which would be dropped in the end io callback of each bio).
5095 * Therefore here we effectively end up waiting for those bios and
5096 * anyone else holding locked ranges without having bumped the inode's
5097 * reference count - if we don't do it, when they access the inode's
5098 * io_tree to unlock a range it may be too late, leading to an
5099 * use-after-free issue.
5100 */
5101 spin_lock(&io_tree->lock);
5102 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5103 struct extent_state *state;
5104 struct extent_state *cached_state = NULL;
5105 u64 start;
5106 u64 end;
5107
5108 node = rb_first(&io_tree->state);
5109 state = rb_entry(node, struct extent_state, rb_node);
5110 start = state->start;
5111 end = state->end;
5112 spin_unlock(&io_tree->lock);
5113
5114 lock_extent_bits(io_tree, start, end, 0, &cached_state);
5115
5116 /*
5117 * If still has DELALLOC flag, the extent didn't reach disk,
5118 * and its reserved space won't be freed by delayed_ref.
5119 * So we need to free its reserved space here.
5120 * (Refer to comment in btrfs_invalidatepage, case 2)
5121 *
5122 * Note, end is the bytenr of last byte, so we need + 1 here.
5123 */
5124 if (state->state & EXTENT_DELALLOC)
5125 btrfs_qgroup_free_data(inode, start, end - start + 1);
5126
5127 clear_extent_bit(io_tree, start, end,
5128 EXTENT_LOCKED | EXTENT_DIRTY |
5129 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5130 EXTENT_DEFRAG, 1, 1,
5131 &cached_state, GFP_NOFS);
5132
5133 cond_resched();
5134 spin_lock(&io_tree->lock);
5135 }
5136 spin_unlock(&io_tree->lock);
5137 }
5138
5139 void btrfs_evict_inode(struct inode *inode)
5140 {
5141 struct btrfs_trans_handle *trans;
5142 struct btrfs_root *root = BTRFS_I(inode)->root;
5143 struct btrfs_block_rsv *rsv, *global_rsv;
5144 int steal_from_global = 0;
5145 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5146 int ret;
5147
5148 trace_btrfs_inode_evict(inode);
5149
5150 evict_inode_truncate_pages(inode);
5151
5152 if (inode->i_nlink &&
5153 ((btrfs_root_refs(&root->root_item) != 0 &&
5154 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5155 btrfs_is_free_space_inode(inode)))
5156 goto no_delete;
5157
5158 if (is_bad_inode(inode)) {
5159 btrfs_orphan_del(NULL, inode);
5160 goto no_delete;
5161 }
5162 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5163 if (!special_file(inode->i_mode))
5164 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5165
5166 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5167
5168 if (root->fs_info->log_root_recovering) {
5169 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5170 &BTRFS_I(inode)->runtime_flags));
5171 goto no_delete;
5172 }
5173
5174 if (inode->i_nlink > 0) {
5175 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5176 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5177 goto no_delete;
5178 }
5179
5180 ret = btrfs_commit_inode_delayed_inode(inode);
5181 if (ret) {
5182 btrfs_orphan_del(NULL, inode);
5183 goto no_delete;
5184 }
5185
5186 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5187 if (!rsv) {
5188 btrfs_orphan_del(NULL, inode);
5189 goto no_delete;
5190 }
5191 rsv->size = min_size;
5192 rsv->failfast = 1;
5193 global_rsv = &root->fs_info->global_block_rsv;
5194
5195 btrfs_i_size_write(inode, 0);
5196
5197 /*
5198 * This is a bit simpler than btrfs_truncate since we've already
5199 * reserved our space for our orphan item in the unlink, so we just
5200 * need to reserve some slack space in case we add bytes and update
5201 * inode item when doing the truncate.
5202 */
5203 while (1) {
5204 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5205 BTRFS_RESERVE_FLUSH_LIMIT);
5206
5207 /*
5208 * Try and steal from the global reserve since we will
5209 * likely not use this space anyway, we want to try as
5210 * hard as possible to get this to work.
5211 */
5212 if (ret)
5213 steal_from_global++;
5214 else
5215 steal_from_global = 0;
5216 ret = 0;
5217
5218 /*
5219 * steal_from_global == 0: we reserved stuff, hooray!
5220 * steal_from_global == 1: we didn't reserve stuff, boo!
5221 * steal_from_global == 2: we've committed, still not a lot of
5222 * room but maybe we'll have room in the global reserve this
5223 * time.
5224 * steal_from_global == 3: abandon all hope!
5225 */
5226 if (steal_from_global > 2) {
5227 btrfs_warn(root->fs_info,
5228 "Could not get space for a delete, will truncate on mount %d",
5229 ret);
5230 btrfs_orphan_del(NULL, inode);
5231 btrfs_free_block_rsv(root, rsv);
5232 goto no_delete;
5233 }
5234
5235 trans = btrfs_join_transaction(root);
5236 if (IS_ERR(trans)) {
5237 btrfs_orphan_del(NULL, inode);
5238 btrfs_free_block_rsv(root, rsv);
5239 goto no_delete;
5240 }
5241
5242 /*
5243 * We can't just steal from the global reserve, we need tomake
5244 * sure there is room to do it, if not we need to commit and try
5245 * again.
5246 */
5247 if (steal_from_global) {
5248 if (!btrfs_check_space_for_delayed_refs(trans, root))
5249 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5250 min_size);
5251 else
5252 ret = -ENOSPC;
5253 }
5254
5255 /*
5256 * Couldn't steal from the global reserve, we have too much
5257 * pending stuff built up, commit the transaction and try it
5258 * again.
5259 */
5260 if (ret) {
5261 ret = btrfs_commit_transaction(trans, root);
5262 if (ret) {
5263 btrfs_orphan_del(NULL, inode);
5264 btrfs_free_block_rsv(root, rsv);
5265 goto no_delete;
5266 }
5267 continue;
5268 } else {
5269 steal_from_global = 0;
5270 }
5271
5272 trans->block_rsv = rsv;
5273
5274 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5275 if (ret != -ENOSPC && ret != -EAGAIN)
5276 break;
5277
5278 trans->block_rsv = &root->fs_info->trans_block_rsv;
5279 btrfs_end_transaction(trans, root);
5280 trans = NULL;
5281 btrfs_btree_balance_dirty(root);
5282 }
5283
5284 btrfs_free_block_rsv(root, rsv);
5285
5286 /*
5287 * Errors here aren't a big deal, it just means we leave orphan items
5288 * in the tree. They will be cleaned up on the next mount.
5289 */
5290 if (ret == 0) {
5291 trans->block_rsv = root->orphan_block_rsv;
5292 btrfs_orphan_del(trans, inode);
5293 } else {
5294 btrfs_orphan_del(NULL, inode);
5295 }
5296
5297 trans->block_rsv = &root->fs_info->trans_block_rsv;
5298 if (!(root == root->fs_info->tree_root ||
5299 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5300 btrfs_return_ino(root, btrfs_ino(inode));
5301
5302 btrfs_end_transaction(trans, root);
5303 btrfs_btree_balance_dirty(root);
5304 no_delete:
5305 btrfs_remove_delayed_node(inode);
5306 clear_inode(inode);
5307 return;
5308 }
5309
5310 /*
5311 * this returns the key found in the dir entry in the location pointer.
5312 * If no dir entries were found, location->objectid is 0.
5313 */
5314 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5315 struct btrfs_key *location)
5316 {
5317 const char *name = dentry->d_name.name;
5318 int namelen = dentry->d_name.len;
5319 struct btrfs_dir_item *di;
5320 struct btrfs_path *path;
5321 struct btrfs_root *root = BTRFS_I(dir)->root;
5322 int ret = 0;
5323
5324 path = btrfs_alloc_path();
5325 if (!path)
5326 return -ENOMEM;
5327
5328 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5329 namelen, 0);
5330 if (IS_ERR(di))
5331 ret = PTR_ERR(di);
5332
5333 if (IS_ERR_OR_NULL(di))
5334 goto out_err;
5335
5336 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5337 out:
5338 btrfs_free_path(path);
5339 return ret;
5340 out_err:
5341 location->objectid = 0;
5342 goto out;
5343 }
5344
5345 /*
5346 * when we hit a tree root in a directory, the btrfs part of the inode
5347 * needs to be changed to reflect the root directory of the tree root. This
5348 * is kind of like crossing a mount point.
5349 */
5350 static int fixup_tree_root_location(struct btrfs_root *root,
5351 struct inode *dir,
5352 struct dentry *dentry,
5353 struct btrfs_key *location,
5354 struct btrfs_root **sub_root)
5355 {
5356 struct btrfs_path *path;
5357 struct btrfs_root *new_root;
5358 struct btrfs_root_ref *ref;
5359 struct extent_buffer *leaf;
5360 struct btrfs_key key;
5361 int ret;
5362 int err = 0;
5363
5364 path = btrfs_alloc_path();
5365 if (!path) {
5366 err = -ENOMEM;
5367 goto out;
5368 }
5369
5370 err = -ENOENT;
5371 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5372 key.type = BTRFS_ROOT_REF_KEY;
5373 key.offset = location->objectid;
5374
5375 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5376 0, 0);
5377 if (ret) {
5378 if (ret < 0)
5379 err = ret;
5380 goto out;
5381 }
5382
5383 leaf = path->nodes[0];
5384 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5385 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5386 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5387 goto out;
5388
5389 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5390 (unsigned long)(ref + 1),
5391 dentry->d_name.len);
5392 if (ret)
5393 goto out;
5394
5395 btrfs_release_path(path);
5396
5397 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5398 if (IS_ERR(new_root)) {
5399 err = PTR_ERR(new_root);
5400 goto out;
5401 }
5402
5403 *sub_root = new_root;
5404 location->objectid = btrfs_root_dirid(&new_root->root_item);
5405 location->type = BTRFS_INODE_ITEM_KEY;
5406 location->offset = 0;
5407 err = 0;
5408 out:
5409 btrfs_free_path(path);
5410 return err;
5411 }
5412
5413 static void inode_tree_add(struct inode *inode)
5414 {
5415 struct btrfs_root *root = BTRFS_I(inode)->root;
5416 struct btrfs_inode *entry;
5417 struct rb_node **p;
5418 struct rb_node *parent;
5419 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5420 u64 ino = btrfs_ino(inode);
5421
5422 if (inode_unhashed(inode))
5423 return;
5424 parent = NULL;
5425 spin_lock(&root->inode_lock);
5426 p = &root->inode_tree.rb_node;
5427 while (*p) {
5428 parent = *p;
5429 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5430
5431 if (ino < btrfs_ino(&entry->vfs_inode))
5432 p = &parent->rb_left;
5433 else if (ino > btrfs_ino(&entry->vfs_inode))
5434 p = &parent->rb_right;
5435 else {
5436 WARN_ON(!(entry->vfs_inode.i_state &
5437 (I_WILL_FREE | I_FREEING)));
5438 rb_replace_node(parent, new, &root->inode_tree);
5439 RB_CLEAR_NODE(parent);
5440 spin_unlock(&root->inode_lock);
5441 return;
5442 }
5443 }
5444 rb_link_node(new, parent, p);
5445 rb_insert_color(new, &root->inode_tree);
5446 spin_unlock(&root->inode_lock);
5447 }
5448
5449 static void inode_tree_del(struct inode *inode)
5450 {
5451 struct btrfs_root *root = BTRFS_I(inode)->root;
5452 int empty = 0;
5453
5454 spin_lock(&root->inode_lock);
5455 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5456 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5457 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5458 empty = RB_EMPTY_ROOT(&root->inode_tree);
5459 }
5460 spin_unlock(&root->inode_lock);
5461
5462 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5463 synchronize_srcu(&root->fs_info->subvol_srcu);
5464 spin_lock(&root->inode_lock);
5465 empty = RB_EMPTY_ROOT(&root->inode_tree);
5466 spin_unlock(&root->inode_lock);
5467 if (empty)
5468 btrfs_add_dead_root(root);
5469 }
5470 }
5471
5472 void btrfs_invalidate_inodes(struct btrfs_root *root)
5473 {
5474 struct rb_node *node;
5475 struct rb_node *prev;
5476 struct btrfs_inode *entry;
5477 struct inode *inode;
5478 u64 objectid = 0;
5479
5480 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5481 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5482
5483 spin_lock(&root->inode_lock);
5484 again:
5485 node = root->inode_tree.rb_node;
5486 prev = NULL;
5487 while (node) {
5488 prev = node;
5489 entry = rb_entry(node, struct btrfs_inode, rb_node);
5490
5491 if (objectid < btrfs_ino(&entry->vfs_inode))
5492 node = node->rb_left;
5493 else if (objectid > btrfs_ino(&entry->vfs_inode))
5494 node = node->rb_right;
5495 else
5496 break;
5497 }
5498 if (!node) {
5499 while (prev) {
5500 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5501 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5502 node = prev;
5503 break;
5504 }
5505 prev = rb_next(prev);
5506 }
5507 }
5508 while (node) {
5509 entry = rb_entry(node, struct btrfs_inode, rb_node);
5510 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5511 inode = igrab(&entry->vfs_inode);
5512 if (inode) {
5513 spin_unlock(&root->inode_lock);
5514 if (atomic_read(&inode->i_count) > 1)
5515 d_prune_aliases(inode);
5516 /*
5517 * btrfs_drop_inode will have it removed from
5518 * the inode cache when its usage count
5519 * hits zero.
5520 */
5521 iput(inode);
5522 cond_resched();
5523 spin_lock(&root->inode_lock);
5524 goto again;
5525 }
5526
5527 if (cond_resched_lock(&root->inode_lock))
5528 goto again;
5529
5530 node = rb_next(node);
5531 }
5532 spin_unlock(&root->inode_lock);
5533 }
5534
5535 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5536 {
5537 struct btrfs_iget_args *args = p;
5538 inode->i_ino = args->location->objectid;
5539 memcpy(&BTRFS_I(inode)->location, args->location,
5540 sizeof(*args->location));
5541 BTRFS_I(inode)->root = args->root;
5542 return 0;
5543 }
5544
5545 static int btrfs_find_actor(struct inode *inode, void *opaque)
5546 {
5547 struct btrfs_iget_args *args = opaque;
5548 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5549 args->root == BTRFS_I(inode)->root;
5550 }
5551
5552 static struct inode *btrfs_iget_locked(struct super_block *s,
5553 struct btrfs_key *location,
5554 struct btrfs_root *root)
5555 {
5556 struct inode *inode;
5557 struct btrfs_iget_args args;
5558 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5559
5560 args.location = location;
5561 args.root = root;
5562
5563 inode = iget5_locked(s, hashval, btrfs_find_actor,
5564 btrfs_init_locked_inode,
5565 (void *)&args);
5566 return inode;
5567 }
5568
5569 /* Get an inode object given its location and corresponding root.
5570 * Returns in *is_new if the inode was read from disk
5571 */
5572 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5573 struct btrfs_root *root, int *new)
5574 {
5575 struct inode *inode;
5576
5577 inode = btrfs_iget_locked(s, location, root);
5578 if (!inode)
5579 return ERR_PTR(-ENOMEM);
5580
5581 if (inode->i_state & I_NEW) {
5582 btrfs_read_locked_inode(inode);
5583 if (!is_bad_inode(inode)) {
5584 inode_tree_add(inode);
5585 unlock_new_inode(inode);
5586 if (new)
5587 *new = 1;
5588 } else {
5589 unlock_new_inode(inode);
5590 iput(inode);
5591 inode = ERR_PTR(-ESTALE);
5592 }
5593 }
5594
5595 return inode;
5596 }
5597
5598 static struct inode *new_simple_dir(struct super_block *s,
5599 struct btrfs_key *key,
5600 struct btrfs_root *root)
5601 {
5602 struct inode *inode = new_inode(s);
5603
5604 if (!inode)
5605 return ERR_PTR(-ENOMEM);
5606
5607 BTRFS_I(inode)->root = root;
5608 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5609 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5610
5611 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5612 inode->i_op = &btrfs_dir_ro_inode_operations;
5613 inode->i_fop = &simple_dir_operations;
5614 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5615 inode->i_mtime = CURRENT_TIME;
5616 inode->i_atime = inode->i_mtime;
5617 inode->i_ctime = inode->i_mtime;
5618 BTRFS_I(inode)->i_otime = inode->i_mtime;
5619
5620 return inode;
5621 }
5622
5623 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5624 {
5625 struct inode *inode;
5626 struct btrfs_root *root = BTRFS_I(dir)->root;
5627 struct btrfs_root *sub_root = root;
5628 struct btrfs_key location;
5629 int index;
5630 int ret = 0;
5631
5632 if (dentry->d_name.len > BTRFS_NAME_LEN)
5633 return ERR_PTR(-ENAMETOOLONG);
5634
5635 ret = btrfs_inode_by_name(dir, dentry, &location);
5636 if (ret < 0)
5637 return ERR_PTR(ret);
5638
5639 if (location.objectid == 0)
5640 return ERR_PTR(-ENOENT);
5641
5642 if (location.type == BTRFS_INODE_ITEM_KEY) {
5643 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5644 return inode;
5645 }
5646
5647 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5648
5649 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5650 ret = fixup_tree_root_location(root, dir, dentry,
5651 &location, &sub_root);
5652 if (ret < 0) {
5653 if (ret != -ENOENT)
5654 inode = ERR_PTR(ret);
5655 else
5656 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5657 } else {
5658 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5659 }
5660 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5661
5662 if (!IS_ERR(inode) && root != sub_root) {
5663 down_read(&root->fs_info->cleanup_work_sem);
5664 if (!(inode->i_sb->s_flags & MS_RDONLY))
5665 ret = btrfs_orphan_cleanup(sub_root);
5666 up_read(&root->fs_info->cleanup_work_sem);
5667 if (ret) {
5668 iput(inode);
5669 inode = ERR_PTR(ret);
5670 }
5671 }
5672
5673 return inode;
5674 }
5675
5676 static int btrfs_dentry_delete(const struct dentry *dentry)
5677 {
5678 struct btrfs_root *root;
5679 struct inode *inode = d_inode(dentry);
5680
5681 if (!inode && !IS_ROOT(dentry))
5682 inode = d_inode(dentry->d_parent);
5683
5684 if (inode) {
5685 root = BTRFS_I(inode)->root;
5686 if (btrfs_root_refs(&root->root_item) == 0)
5687 return 1;
5688
5689 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5690 return 1;
5691 }
5692 return 0;
5693 }
5694
5695 static void btrfs_dentry_release(struct dentry *dentry)
5696 {
5697 kfree(dentry->d_fsdata);
5698 }
5699
5700 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5701 unsigned int flags)
5702 {
5703 struct inode *inode;
5704
5705 inode = btrfs_lookup_dentry(dir, dentry);
5706 if (IS_ERR(inode)) {
5707 if (PTR_ERR(inode) == -ENOENT)
5708 inode = NULL;
5709 else
5710 return ERR_CAST(inode);
5711 }
5712
5713 return d_splice_alias(inode, dentry);
5714 }
5715
5716 unsigned char btrfs_filetype_table[] = {
5717 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5718 };
5719
5720 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5721 {
5722 struct inode *inode = file_inode(file);
5723 struct btrfs_root *root = BTRFS_I(inode)->root;
5724 struct btrfs_item *item;
5725 struct btrfs_dir_item *di;
5726 struct btrfs_key key;
5727 struct btrfs_key found_key;
5728 struct btrfs_path *path;
5729 struct list_head ins_list;
5730 struct list_head del_list;
5731 int ret;
5732 struct extent_buffer *leaf;
5733 int slot;
5734 unsigned char d_type;
5735 int over = 0;
5736 u32 di_cur;
5737 u32 di_total;
5738 u32 di_len;
5739 int key_type = BTRFS_DIR_INDEX_KEY;
5740 char tmp_name[32];
5741 char *name_ptr;
5742 int name_len;
5743 int is_curr = 0; /* ctx->pos points to the current index? */
5744
5745 /* FIXME, use a real flag for deciding about the key type */
5746 if (root->fs_info->tree_root == root)
5747 key_type = BTRFS_DIR_ITEM_KEY;
5748
5749 if (!dir_emit_dots(file, ctx))
5750 return 0;
5751
5752 path = btrfs_alloc_path();
5753 if (!path)
5754 return -ENOMEM;
5755
5756 path->reada = 1;
5757
5758 if (key_type == BTRFS_DIR_INDEX_KEY) {
5759 INIT_LIST_HEAD(&ins_list);
5760 INIT_LIST_HEAD(&del_list);
5761 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5762 }
5763
5764 key.type = key_type;
5765 key.offset = ctx->pos;
5766 key.objectid = btrfs_ino(inode);
5767
5768 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5769 if (ret < 0)
5770 goto err;
5771
5772 while (1) {
5773 leaf = path->nodes[0];
5774 slot = path->slots[0];
5775 if (slot >= btrfs_header_nritems(leaf)) {
5776 ret = btrfs_next_leaf(root, path);
5777 if (ret < 0)
5778 goto err;
5779 else if (ret > 0)
5780 break;
5781 continue;
5782 }
5783
5784 item = btrfs_item_nr(slot);
5785 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5786
5787 if (found_key.objectid != key.objectid)
5788 break;
5789 if (found_key.type != key_type)
5790 break;
5791 if (found_key.offset < ctx->pos)
5792 goto next;
5793 if (key_type == BTRFS_DIR_INDEX_KEY &&
5794 btrfs_should_delete_dir_index(&del_list,
5795 found_key.offset))
5796 goto next;
5797
5798 ctx->pos = found_key.offset;
5799 is_curr = 1;
5800
5801 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5802 di_cur = 0;
5803 di_total = btrfs_item_size(leaf, item);
5804
5805 while (di_cur < di_total) {
5806 struct btrfs_key location;
5807
5808 if (verify_dir_item(root, leaf, di))
5809 break;
5810
5811 name_len = btrfs_dir_name_len(leaf, di);
5812 if (name_len <= sizeof(tmp_name)) {
5813 name_ptr = tmp_name;
5814 } else {
5815 name_ptr = kmalloc(name_len, GFP_NOFS);
5816 if (!name_ptr) {
5817 ret = -ENOMEM;
5818 goto err;
5819 }
5820 }
5821 read_extent_buffer(leaf, name_ptr,
5822 (unsigned long)(di + 1), name_len);
5823
5824 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5825 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5826
5827
5828 /* is this a reference to our own snapshot? If so
5829 * skip it.
5830 *
5831 * In contrast to old kernels, we insert the snapshot's
5832 * dir item and dir index after it has been created, so
5833 * we won't find a reference to our own snapshot. We
5834 * still keep the following code for backward
5835 * compatibility.
5836 */
5837 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5838 location.objectid == root->root_key.objectid) {
5839 over = 0;
5840 goto skip;
5841 }
5842 over = !dir_emit(ctx, name_ptr, name_len,
5843 location.objectid, d_type);
5844
5845 skip:
5846 if (name_ptr != tmp_name)
5847 kfree(name_ptr);
5848
5849 if (over)
5850 goto nopos;
5851 di_len = btrfs_dir_name_len(leaf, di) +
5852 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5853 di_cur += di_len;
5854 di = (struct btrfs_dir_item *)((char *)di + di_len);
5855 }
5856 next:
5857 path->slots[0]++;
5858 }
5859
5860 if (key_type == BTRFS_DIR_INDEX_KEY) {
5861 if (is_curr)
5862 ctx->pos++;
5863 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5864 if (ret)
5865 goto nopos;
5866 }
5867
5868 /* Reached end of directory/root. Bump pos past the last item. */
5869 ctx->pos++;
5870
5871 /*
5872 * Stop new entries from being returned after we return the last
5873 * entry.
5874 *
5875 * New directory entries are assigned a strictly increasing
5876 * offset. This means that new entries created during readdir
5877 * are *guaranteed* to be seen in the future by that readdir.
5878 * This has broken buggy programs which operate on names as
5879 * they're returned by readdir. Until we re-use freed offsets
5880 * we have this hack to stop new entries from being returned
5881 * under the assumption that they'll never reach this huge
5882 * offset.
5883 *
5884 * This is being careful not to overflow 32bit loff_t unless the
5885 * last entry requires it because doing so has broken 32bit apps
5886 * in the past.
5887 */
5888 if (key_type == BTRFS_DIR_INDEX_KEY) {
5889 if (ctx->pos >= INT_MAX)
5890 ctx->pos = LLONG_MAX;
5891 else
5892 ctx->pos = INT_MAX;
5893 }
5894 nopos:
5895 ret = 0;
5896 err:
5897 if (key_type == BTRFS_DIR_INDEX_KEY)
5898 btrfs_put_delayed_items(&ins_list, &del_list);
5899 btrfs_free_path(path);
5900 return ret;
5901 }
5902
5903 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5904 {
5905 struct btrfs_root *root = BTRFS_I(inode)->root;
5906 struct btrfs_trans_handle *trans;
5907 int ret = 0;
5908 bool nolock = false;
5909
5910 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5911 return 0;
5912
5913 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5914 nolock = true;
5915
5916 if (wbc->sync_mode == WB_SYNC_ALL) {
5917 if (nolock)
5918 trans = btrfs_join_transaction_nolock(root);
5919 else
5920 trans = btrfs_join_transaction(root);
5921 if (IS_ERR(trans))
5922 return PTR_ERR(trans);
5923 ret = btrfs_commit_transaction(trans, root);
5924 }
5925 return ret;
5926 }
5927
5928 /*
5929 * This is somewhat expensive, updating the tree every time the
5930 * inode changes. But, it is most likely to find the inode in cache.
5931 * FIXME, needs more benchmarking...there are no reasons other than performance
5932 * to keep or drop this code.
5933 */
5934 static int btrfs_dirty_inode(struct inode *inode)
5935 {
5936 struct btrfs_root *root = BTRFS_I(inode)->root;
5937 struct btrfs_trans_handle *trans;
5938 int ret;
5939
5940 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5941 return 0;
5942
5943 trans = btrfs_join_transaction(root);
5944 if (IS_ERR(trans))
5945 return PTR_ERR(trans);
5946
5947 ret = btrfs_update_inode(trans, root, inode);
5948 if (ret && ret == -ENOSPC) {
5949 /* whoops, lets try again with the full transaction */
5950 btrfs_end_transaction(trans, root);
5951 trans = btrfs_start_transaction(root, 1);
5952 if (IS_ERR(trans))
5953 return PTR_ERR(trans);
5954
5955 ret = btrfs_update_inode(trans, root, inode);
5956 }
5957 btrfs_end_transaction(trans, root);
5958 if (BTRFS_I(inode)->delayed_node)
5959 btrfs_balance_delayed_items(root);
5960
5961 return ret;
5962 }
5963
5964 /*
5965 * This is a copy of file_update_time. We need this so we can return error on
5966 * ENOSPC for updating the inode in the case of file write and mmap writes.
5967 */
5968 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5969 int flags)
5970 {
5971 struct btrfs_root *root = BTRFS_I(inode)->root;
5972
5973 if (btrfs_root_readonly(root))
5974 return -EROFS;
5975
5976 if (flags & S_VERSION)
5977 inode_inc_iversion(inode);
5978 if (flags & S_CTIME)
5979 inode->i_ctime = *now;
5980 if (flags & S_MTIME)
5981 inode->i_mtime = *now;
5982 if (flags & S_ATIME)
5983 inode->i_atime = *now;
5984 return btrfs_dirty_inode(inode);
5985 }
5986
5987 /*
5988 * find the highest existing sequence number in a directory
5989 * and then set the in-memory index_cnt variable to reflect
5990 * free sequence numbers
5991 */
5992 static int btrfs_set_inode_index_count(struct inode *inode)
5993 {
5994 struct btrfs_root *root = BTRFS_I(inode)->root;
5995 struct btrfs_key key, found_key;
5996 struct btrfs_path *path;
5997 struct extent_buffer *leaf;
5998 int ret;
5999
6000 key.objectid = btrfs_ino(inode);
6001 key.type = BTRFS_DIR_INDEX_KEY;
6002 key.offset = (u64)-1;
6003
6004 path = btrfs_alloc_path();
6005 if (!path)
6006 return -ENOMEM;
6007
6008 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6009 if (ret < 0)
6010 goto out;
6011 /* FIXME: we should be able to handle this */
6012 if (ret == 0)
6013 goto out;
6014 ret = 0;
6015
6016 /*
6017 * MAGIC NUMBER EXPLANATION:
6018 * since we search a directory based on f_pos we have to start at 2
6019 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6020 * else has to start at 2
6021 */
6022 if (path->slots[0] == 0) {
6023 BTRFS_I(inode)->index_cnt = 2;
6024 goto out;
6025 }
6026
6027 path->slots[0]--;
6028
6029 leaf = path->nodes[0];
6030 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6031
6032 if (found_key.objectid != btrfs_ino(inode) ||
6033 found_key.type != BTRFS_DIR_INDEX_KEY) {
6034 BTRFS_I(inode)->index_cnt = 2;
6035 goto out;
6036 }
6037
6038 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6039 out:
6040 btrfs_free_path(path);
6041 return ret;
6042 }
6043
6044 /*
6045 * helper to find a free sequence number in a given directory. This current
6046 * code is very simple, later versions will do smarter things in the btree
6047 */
6048 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6049 {
6050 int ret = 0;
6051
6052 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6053 ret = btrfs_inode_delayed_dir_index_count(dir);
6054 if (ret) {
6055 ret = btrfs_set_inode_index_count(dir);
6056 if (ret)
6057 return ret;
6058 }
6059 }
6060
6061 *index = BTRFS_I(dir)->index_cnt;
6062 BTRFS_I(dir)->index_cnt++;
6063
6064 return ret;
6065 }
6066
6067 static int btrfs_insert_inode_locked(struct inode *inode)
6068 {
6069 struct btrfs_iget_args args;
6070 args.location = &BTRFS_I(inode)->location;
6071 args.root = BTRFS_I(inode)->root;
6072
6073 return insert_inode_locked4(inode,
6074 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6075 btrfs_find_actor, &args);
6076 }
6077
6078 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6079 struct btrfs_root *root,
6080 struct inode *dir,
6081 const char *name, int name_len,
6082 u64 ref_objectid, u64 objectid,
6083 umode_t mode, u64 *index)
6084 {
6085 struct inode *inode;
6086 struct btrfs_inode_item *inode_item;
6087 struct btrfs_key *location;
6088 struct btrfs_path *path;
6089 struct btrfs_inode_ref *ref;
6090 struct btrfs_key key[2];
6091 u32 sizes[2];
6092 int nitems = name ? 2 : 1;
6093 unsigned long ptr;
6094 int ret;
6095
6096 path = btrfs_alloc_path();
6097 if (!path)
6098 return ERR_PTR(-ENOMEM);
6099
6100 inode = new_inode(root->fs_info->sb);
6101 if (!inode) {
6102 btrfs_free_path(path);
6103 return ERR_PTR(-ENOMEM);
6104 }
6105
6106 /*
6107 * O_TMPFILE, set link count to 0, so that after this point,
6108 * we fill in an inode item with the correct link count.
6109 */
6110 if (!name)
6111 set_nlink(inode, 0);
6112
6113 /*
6114 * we have to initialize this early, so we can reclaim the inode
6115 * number if we fail afterwards in this function.
6116 */
6117 inode->i_ino = objectid;
6118
6119 if (dir && name) {
6120 trace_btrfs_inode_request(dir);
6121
6122 ret = btrfs_set_inode_index(dir, index);
6123 if (ret) {
6124 btrfs_free_path(path);
6125 iput(inode);
6126 return ERR_PTR(ret);
6127 }
6128 } else if (dir) {
6129 *index = 0;
6130 }
6131 /*
6132 * index_cnt is ignored for everything but a dir,
6133 * btrfs_get_inode_index_count has an explanation for the magic
6134 * number
6135 */
6136 BTRFS_I(inode)->index_cnt = 2;
6137 BTRFS_I(inode)->dir_index = *index;
6138 BTRFS_I(inode)->root = root;
6139 BTRFS_I(inode)->generation = trans->transid;
6140 inode->i_generation = BTRFS_I(inode)->generation;
6141
6142 /*
6143 * We could have gotten an inode number from somebody who was fsynced
6144 * and then removed in this same transaction, so let's just set full
6145 * sync since it will be a full sync anyway and this will blow away the
6146 * old info in the log.
6147 */
6148 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6149
6150 key[0].objectid = objectid;
6151 key[0].type = BTRFS_INODE_ITEM_KEY;
6152 key[0].offset = 0;
6153
6154 sizes[0] = sizeof(struct btrfs_inode_item);
6155
6156 if (name) {
6157 /*
6158 * Start new inodes with an inode_ref. This is slightly more
6159 * efficient for small numbers of hard links since they will
6160 * be packed into one item. Extended refs will kick in if we
6161 * add more hard links than can fit in the ref item.
6162 */
6163 key[1].objectid = objectid;
6164 key[1].type = BTRFS_INODE_REF_KEY;
6165 key[1].offset = ref_objectid;
6166
6167 sizes[1] = name_len + sizeof(*ref);
6168 }
6169
6170 location = &BTRFS_I(inode)->location;
6171 location->objectid = objectid;
6172 location->offset = 0;
6173 location->type = BTRFS_INODE_ITEM_KEY;
6174
6175 ret = btrfs_insert_inode_locked(inode);
6176 if (ret < 0)
6177 goto fail;
6178
6179 path->leave_spinning = 1;
6180 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6181 if (ret != 0)
6182 goto fail_unlock;
6183
6184 inode_init_owner(inode, dir, mode);
6185 inode_set_bytes(inode, 0);
6186
6187 inode->i_mtime = CURRENT_TIME;
6188 inode->i_atime = inode->i_mtime;
6189 inode->i_ctime = inode->i_mtime;
6190 BTRFS_I(inode)->i_otime = inode->i_mtime;
6191
6192 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6193 struct btrfs_inode_item);
6194 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6195 sizeof(*inode_item));
6196 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6197
6198 if (name) {
6199 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6200 struct btrfs_inode_ref);
6201 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6202 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6203 ptr = (unsigned long)(ref + 1);
6204 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6205 }
6206
6207 btrfs_mark_buffer_dirty(path->nodes[0]);
6208 btrfs_free_path(path);
6209
6210 btrfs_inherit_iflags(inode, dir);
6211
6212 if (S_ISREG(mode)) {
6213 if (btrfs_test_opt(root, NODATASUM))
6214 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6215 if (btrfs_test_opt(root, NODATACOW))
6216 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6217 BTRFS_INODE_NODATASUM;
6218 }
6219
6220 inode_tree_add(inode);
6221
6222 trace_btrfs_inode_new(inode);
6223 btrfs_set_inode_last_trans(trans, inode);
6224
6225 btrfs_update_root_times(trans, root);
6226
6227 ret = btrfs_inode_inherit_props(trans, inode, dir);
6228 if (ret)
6229 btrfs_err(root->fs_info,
6230 "error inheriting props for ino %llu (root %llu): %d",
6231 btrfs_ino(inode), root->root_key.objectid, ret);
6232
6233 return inode;
6234
6235 fail_unlock:
6236 unlock_new_inode(inode);
6237 fail:
6238 if (dir && name)
6239 BTRFS_I(dir)->index_cnt--;
6240 btrfs_free_path(path);
6241 iput(inode);
6242 return ERR_PTR(ret);
6243 }
6244
6245 static inline u8 btrfs_inode_type(struct inode *inode)
6246 {
6247 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6248 }
6249
6250 /*
6251 * utility function to add 'inode' into 'parent_inode' with
6252 * a give name and a given sequence number.
6253 * if 'add_backref' is true, also insert a backref from the
6254 * inode to the parent directory.
6255 */
6256 int btrfs_add_link(struct btrfs_trans_handle *trans,
6257 struct inode *parent_inode, struct inode *inode,
6258 const char *name, int name_len, int add_backref, u64 index)
6259 {
6260 int ret = 0;
6261 struct btrfs_key key;
6262 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6263 u64 ino = btrfs_ino(inode);
6264 u64 parent_ino = btrfs_ino(parent_inode);
6265
6266 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6267 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6268 } else {
6269 key.objectid = ino;
6270 key.type = BTRFS_INODE_ITEM_KEY;
6271 key.offset = 0;
6272 }
6273
6274 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6275 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6276 key.objectid, root->root_key.objectid,
6277 parent_ino, index, name, name_len);
6278 } else if (add_backref) {
6279 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6280 parent_ino, index);
6281 }
6282
6283 /* Nothing to clean up yet */
6284 if (ret)
6285 return ret;
6286
6287 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6288 parent_inode, &key,
6289 btrfs_inode_type(inode), index);
6290 if (ret == -EEXIST || ret == -EOVERFLOW)
6291 goto fail_dir_item;
6292 else if (ret) {
6293 btrfs_abort_transaction(trans, root, ret);
6294 return ret;
6295 }
6296
6297 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6298 name_len * 2);
6299 inode_inc_iversion(parent_inode);
6300 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6301 ret = btrfs_update_inode(trans, root, parent_inode);
6302 if (ret)
6303 btrfs_abort_transaction(trans, root, ret);
6304 return ret;
6305
6306 fail_dir_item:
6307 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6308 u64 local_index;
6309 int err;
6310 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6311 key.objectid, root->root_key.objectid,
6312 parent_ino, &local_index, name, name_len);
6313
6314 } else if (add_backref) {
6315 u64 local_index;
6316 int err;
6317
6318 err = btrfs_del_inode_ref(trans, root, name, name_len,
6319 ino, parent_ino, &local_index);
6320 }
6321 return ret;
6322 }
6323
6324 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6325 struct inode *dir, struct dentry *dentry,
6326 struct inode *inode, int backref, u64 index)
6327 {
6328 int err = btrfs_add_link(trans, dir, inode,
6329 dentry->d_name.name, dentry->d_name.len,
6330 backref, index);
6331 if (err > 0)
6332 err = -EEXIST;
6333 return err;
6334 }
6335
6336 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6337 umode_t mode, dev_t rdev)
6338 {
6339 struct btrfs_trans_handle *trans;
6340 struct btrfs_root *root = BTRFS_I(dir)->root;
6341 struct inode *inode = NULL;
6342 int err;
6343 int drop_inode = 0;
6344 u64 objectid;
6345 u64 index = 0;
6346
6347 /*
6348 * 2 for inode item and ref
6349 * 2 for dir items
6350 * 1 for xattr if selinux is on
6351 */
6352 trans = btrfs_start_transaction(root, 5);
6353 if (IS_ERR(trans))
6354 return PTR_ERR(trans);
6355
6356 err = btrfs_find_free_ino(root, &objectid);
6357 if (err)
6358 goto out_unlock;
6359
6360 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6361 dentry->d_name.len, btrfs_ino(dir), objectid,
6362 mode, &index);
6363 if (IS_ERR(inode)) {
6364 err = PTR_ERR(inode);
6365 goto out_unlock;
6366 }
6367
6368 /*
6369 * If the active LSM wants to access the inode during
6370 * d_instantiate it needs these. Smack checks to see
6371 * if the filesystem supports xattrs by looking at the
6372 * ops vector.
6373 */
6374 inode->i_op = &btrfs_special_inode_operations;
6375 init_special_inode(inode, inode->i_mode, rdev);
6376
6377 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6378 if (err)
6379 goto out_unlock_inode;
6380
6381 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6382 if (err) {
6383 goto out_unlock_inode;
6384 } else {
6385 btrfs_update_inode(trans, root, inode);
6386 unlock_new_inode(inode);
6387 d_instantiate(dentry, inode);
6388 }
6389
6390 out_unlock:
6391 btrfs_end_transaction(trans, root);
6392 btrfs_balance_delayed_items(root);
6393 btrfs_btree_balance_dirty(root);
6394 if (drop_inode) {
6395 inode_dec_link_count(inode);
6396 iput(inode);
6397 }
6398 return err;
6399
6400 out_unlock_inode:
6401 drop_inode = 1;
6402 unlock_new_inode(inode);
6403 goto out_unlock;
6404
6405 }
6406
6407 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6408 umode_t mode, bool excl)
6409 {
6410 struct btrfs_trans_handle *trans;
6411 struct btrfs_root *root = BTRFS_I(dir)->root;
6412 struct inode *inode = NULL;
6413 int drop_inode_on_err = 0;
6414 int err;
6415 u64 objectid;
6416 u64 index = 0;
6417
6418 /*
6419 * 2 for inode item and ref
6420 * 2 for dir items
6421 * 1 for xattr if selinux is on
6422 */
6423 trans = btrfs_start_transaction(root, 5);
6424 if (IS_ERR(trans))
6425 return PTR_ERR(trans);
6426
6427 err = btrfs_find_free_ino(root, &objectid);
6428 if (err)
6429 goto out_unlock;
6430
6431 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6432 dentry->d_name.len, btrfs_ino(dir), objectid,
6433 mode, &index);
6434 if (IS_ERR(inode)) {
6435 err = PTR_ERR(inode);
6436 goto out_unlock;
6437 }
6438 drop_inode_on_err = 1;
6439 /*
6440 * If the active LSM wants to access the inode during
6441 * d_instantiate it needs these. Smack checks to see
6442 * if the filesystem supports xattrs by looking at the
6443 * ops vector.
6444 */
6445 inode->i_fop = &btrfs_file_operations;
6446 inode->i_op = &btrfs_file_inode_operations;
6447 inode->i_mapping->a_ops = &btrfs_aops;
6448
6449 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6450 if (err)
6451 goto out_unlock_inode;
6452
6453 err = btrfs_update_inode(trans, root, inode);
6454 if (err)
6455 goto out_unlock_inode;
6456
6457 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6458 if (err)
6459 goto out_unlock_inode;
6460
6461 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6462 unlock_new_inode(inode);
6463 d_instantiate(dentry, inode);
6464
6465 out_unlock:
6466 btrfs_end_transaction(trans, root);
6467 if (err && drop_inode_on_err) {
6468 inode_dec_link_count(inode);
6469 iput(inode);
6470 }
6471 btrfs_balance_delayed_items(root);
6472 btrfs_btree_balance_dirty(root);
6473 return err;
6474
6475 out_unlock_inode:
6476 unlock_new_inode(inode);
6477 goto out_unlock;
6478
6479 }
6480
6481 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6482 struct dentry *dentry)
6483 {
6484 struct btrfs_trans_handle *trans;
6485 struct btrfs_root *root = BTRFS_I(dir)->root;
6486 struct inode *inode = d_inode(old_dentry);
6487 u64 index;
6488 int err;
6489 int drop_inode = 0;
6490
6491 /* do not allow sys_link's with other subvols of the same device */
6492 if (root->objectid != BTRFS_I(inode)->root->objectid)
6493 return -EXDEV;
6494
6495 if (inode->i_nlink >= BTRFS_LINK_MAX)
6496 return -EMLINK;
6497
6498 err = btrfs_set_inode_index(dir, &index);
6499 if (err)
6500 goto fail;
6501
6502 /*
6503 * 2 items for inode and inode ref
6504 * 2 items for dir items
6505 * 1 item for parent inode
6506 */
6507 trans = btrfs_start_transaction(root, 5);
6508 if (IS_ERR(trans)) {
6509 err = PTR_ERR(trans);
6510 goto fail;
6511 }
6512
6513 /* There are several dir indexes for this inode, clear the cache. */
6514 BTRFS_I(inode)->dir_index = 0ULL;
6515 inc_nlink(inode);
6516 inode_inc_iversion(inode);
6517 inode->i_ctime = CURRENT_TIME;
6518 ihold(inode);
6519 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6520
6521 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6522
6523 if (err) {
6524 drop_inode = 1;
6525 } else {
6526 struct dentry *parent = dentry->d_parent;
6527 err = btrfs_update_inode(trans, root, inode);
6528 if (err)
6529 goto fail;
6530 if (inode->i_nlink == 1) {
6531 /*
6532 * If new hard link count is 1, it's a file created
6533 * with open(2) O_TMPFILE flag.
6534 */
6535 err = btrfs_orphan_del(trans, inode);
6536 if (err)
6537 goto fail;
6538 }
6539 d_instantiate(dentry, inode);
6540 btrfs_log_new_name(trans, inode, NULL, parent);
6541 }
6542
6543 btrfs_end_transaction(trans, root);
6544 btrfs_balance_delayed_items(root);
6545 fail:
6546 if (drop_inode) {
6547 inode_dec_link_count(inode);
6548 iput(inode);
6549 }
6550 btrfs_btree_balance_dirty(root);
6551 return err;
6552 }
6553
6554 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6555 {
6556 struct inode *inode = NULL;
6557 struct btrfs_trans_handle *trans;
6558 struct btrfs_root *root = BTRFS_I(dir)->root;
6559 int err = 0;
6560 int drop_on_err = 0;
6561 u64 objectid = 0;
6562 u64 index = 0;
6563
6564 /*
6565 * 2 items for inode and ref
6566 * 2 items for dir items
6567 * 1 for xattr if selinux is on
6568 */
6569 trans = btrfs_start_transaction(root, 5);
6570 if (IS_ERR(trans))
6571 return PTR_ERR(trans);
6572
6573 err = btrfs_find_free_ino(root, &objectid);
6574 if (err)
6575 goto out_fail;
6576
6577 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6578 dentry->d_name.len, btrfs_ino(dir), objectid,
6579 S_IFDIR | mode, &index);
6580 if (IS_ERR(inode)) {
6581 err = PTR_ERR(inode);
6582 goto out_fail;
6583 }
6584
6585 drop_on_err = 1;
6586 /* these must be set before we unlock the inode */
6587 inode->i_op = &btrfs_dir_inode_operations;
6588 inode->i_fop = &btrfs_dir_file_operations;
6589
6590 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6591 if (err)
6592 goto out_fail_inode;
6593
6594 btrfs_i_size_write(inode, 0);
6595 err = btrfs_update_inode(trans, root, inode);
6596 if (err)
6597 goto out_fail_inode;
6598
6599 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6600 dentry->d_name.len, 0, index);
6601 if (err)
6602 goto out_fail_inode;
6603
6604 d_instantiate(dentry, inode);
6605 /*
6606 * mkdir is special. We're unlocking after we call d_instantiate
6607 * to avoid a race with nfsd calling d_instantiate.
6608 */
6609 unlock_new_inode(inode);
6610 drop_on_err = 0;
6611
6612 out_fail:
6613 btrfs_end_transaction(trans, root);
6614 if (drop_on_err) {
6615 inode_dec_link_count(inode);
6616 iput(inode);
6617 }
6618 btrfs_balance_delayed_items(root);
6619 btrfs_btree_balance_dirty(root);
6620 return err;
6621
6622 out_fail_inode:
6623 unlock_new_inode(inode);
6624 goto out_fail;
6625 }
6626
6627 /* Find next extent map of a given extent map, caller needs to ensure locks */
6628 static struct extent_map *next_extent_map(struct extent_map *em)
6629 {
6630 struct rb_node *next;
6631
6632 next = rb_next(&em->rb_node);
6633 if (!next)
6634 return NULL;
6635 return container_of(next, struct extent_map, rb_node);
6636 }
6637
6638 static struct extent_map *prev_extent_map(struct extent_map *em)
6639 {
6640 struct rb_node *prev;
6641
6642 prev = rb_prev(&em->rb_node);
6643 if (!prev)
6644 return NULL;
6645 return container_of(prev, struct extent_map, rb_node);
6646 }
6647
6648 /* helper for btfs_get_extent. Given an existing extent in the tree,
6649 * the existing extent is the nearest extent to map_start,
6650 * and an extent that you want to insert, deal with overlap and insert
6651 * the best fitted new extent into the tree.
6652 */
6653 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6654 struct extent_map *existing,
6655 struct extent_map *em,
6656 u64 map_start)
6657 {
6658 struct extent_map *prev;
6659 struct extent_map *next;
6660 u64 start;
6661 u64 end;
6662 u64 start_diff;
6663
6664 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6665
6666 if (existing->start > map_start) {
6667 next = existing;
6668 prev = prev_extent_map(next);
6669 } else {
6670 prev = existing;
6671 next = next_extent_map(prev);
6672 }
6673
6674 start = prev ? extent_map_end(prev) : em->start;
6675 start = max_t(u64, start, em->start);
6676 end = next ? next->start : extent_map_end(em);
6677 end = min_t(u64, end, extent_map_end(em));
6678 start_diff = start - em->start;
6679 em->start = start;
6680 em->len = end - start;
6681 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6682 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6683 em->block_start += start_diff;
6684 em->block_len -= start_diff;
6685 }
6686 return add_extent_mapping(em_tree, em, 0);
6687 }
6688
6689 static noinline int uncompress_inline(struct btrfs_path *path,
6690 struct inode *inode, struct page *page,
6691 size_t pg_offset, u64 extent_offset,
6692 struct btrfs_file_extent_item *item)
6693 {
6694 int ret;
6695 struct extent_buffer *leaf = path->nodes[0];
6696 char *tmp;
6697 size_t max_size;
6698 unsigned long inline_size;
6699 unsigned long ptr;
6700 int compress_type;
6701
6702 WARN_ON(pg_offset != 0);
6703 compress_type = btrfs_file_extent_compression(leaf, item);
6704 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6705 inline_size = btrfs_file_extent_inline_item_len(leaf,
6706 btrfs_item_nr(path->slots[0]));
6707 tmp = kmalloc(inline_size, GFP_NOFS);
6708 if (!tmp)
6709 return -ENOMEM;
6710 ptr = btrfs_file_extent_inline_start(item);
6711
6712 read_extent_buffer(leaf, tmp, ptr, inline_size);
6713
6714 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6715 ret = btrfs_decompress(compress_type, tmp, page,
6716 extent_offset, inline_size, max_size);
6717 kfree(tmp);
6718 return ret;
6719 }
6720
6721 /*
6722 * a bit scary, this does extent mapping from logical file offset to the disk.
6723 * the ugly parts come from merging extents from the disk with the in-ram
6724 * representation. This gets more complex because of the data=ordered code,
6725 * where the in-ram extents might be locked pending data=ordered completion.
6726 *
6727 * This also copies inline extents directly into the page.
6728 */
6729
6730 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6731 size_t pg_offset, u64 start, u64 len,
6732 int create)
6733 {
6734 int ret;
6735 int err = 0;
6736 u64 extent_start = 0;
6737 u64 extent_end = 0;
6738 u64 objectid = btrfs_ino(inode);
6739 u32 found_type;
6740 struct btrfs_path *path = NULL;
6741 struct btrfs_root *root = BTRFS_I(inode)->root;
6742 struct btrfs_file_extent_item *item;
6743 struct extent_buffer *leaf;
6744 struct btrfs_key found_key;
6745 struct extent_map *em = NULL;
6746 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6747 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6748 struct btrfs_trans_handle *trans = NULL;
6749 const bool new_inline = !page || create;
6750
6751 again:
6752 read_lock(&em_tree->lock);
6753 em = lookup_extent_mapping(em_tree, start, len);
6754 if (em)
6755 em->bdev = root->fs_info->fs_devices->latest_bdev;
6756 read_unlock(&em_tree->lock);
6757
6758 if (em) {
6759 if (em->start > start || em->start + em->len <= start)
6760 free_extent_map(em);
6761 else if (em->block_start == EXTENT_MAP_INLINE && page)
6762 free_extent_map(em);
6763 else
6764 goto out;
6765 }
6766 em = alloc_extent_map();
6767 if (!em) {
6768 err = -ENOMEM;
6769 goto out;
6770 }
6771 em->bdev = root->fs_info->fs_devices->latest_bdev;
6772 em->start = EXTENT_MAP_HOLE;
6773 em->orig_start = EXTENT_MAP_HOLE;
6774 em->len = (u64)-1;
6775 em->block_len = (u64)-1;
6776
6777 if (!path) {
6778 path = btrfs_alloc_path();
6779 if (!path) {
6780 err = -ENOMEM;
6781 goto out;
6782 }
6783 /*
6784 * Chances are we'll be called again, so go ahead and do
6785 * readahead
6786 */
6787 path->reada = 1;
6788 }
6789
6790 ret = btrfs_lookup_file_extent(trans, root, path,
6791 objectid, start, trans != NULL);
6792 if (ret < 0) {
6793 err = ret;
6794 goto out;
6795 }
6796
6797 if (ret != 0) {
6798 if (path->slots[0] == 0)
6799 goto not_found;
6800 path->slots[0]--;
6801 }
6802
6803 leaf = path->nodes[0];
6804 item = btrfs_item_ptr(leaf, path->slots[0],
6805 struct btrfs_file_extent_item);
6806 /* are we inside the extent that was found? */
6807 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6808 found_type = found_key.type;
6809 if (found_key.objectid != objectid ||
6810 found_type != BTRFS_EXTENT_DATA_KEY) {
6811 /*
6812 * If we backup past the first extent we want to move forward
6813 * and see if there is an extent in front of us, otherwise we'll
6814 * say there is a hole for our whole search range which can
6815 * cause problems.
6816 */
6817 extent_end = start;
6818 goto next;
6819 }
6820
6821 found_type = btrfs_file_extent_type(leaf, item);
6822 extent_start = found_key.offset;
6823 if (found_type == BTRFS_FILE_EXTENT_REG ||
6824 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6825 extent_end = extent_start +
6826 btrfs_file_extent_num_bytes(leaf, item);
6827 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6828 size_t size;
6829 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6830 extent_end = ALIGN(extent_start + size, root->sectorsize);
6831 }
6832 next:
6833 if (start >= extent_end) {
6834 path->slots[0]++;
6835 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6836 ret = btrfs_next_leaf(root, path);
6837 if (ret < 0) {
6838 err = ret;
6839 goto out;
6840 }
6841 if (ret > 0)
6842 goto not_found;
6843 leaf = path->nodes[0];
6844 }
6845 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6846 if (found_key.objectid != objectid ||
6847 found_key.type != BTRFS_EXTENT_DATA_KEY)
6848 goto not_found;
6849 if (start + len <= found_key.offset)
6850 goto not_found;
6851 if (start > found_key.offset)
6852 goto next;
6853 em->start = start;
6854 em->orig_start = start;
6855 em->len = found_key.offset - start;
6856 goto not_found_em;
6857 }
6858
6859 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6860
6861 if (found_type == BTRFS_FILE_EXTENT_REG ||
6862 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6863 goto insert;
6864 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6865 unsigned long ptr;
6866 char *map;
6867 size_t size;
6868 size_t extent_offset;
6869 size_t copy_size;
6870
6871 if (new_inline)
6872 goto out;
6873
6874 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6875 extent_offset = page_offset(page) + pg_offset - extent_start;
6876 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6877 size - extent_offset);
6878 em->start = extent_start + extent_offset;
6879 em->len = ALIGN(copy_size, root->sectorsize);
6880 em->orig_block_len = em->len;
6881 em->orig_start = em->start;
6882 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6883 if (create == 0 && !PageUptodate(page)) {
6884 if (btrfs_file_extent_compression(leaf, item) !=
6885 BTRFS_COMPRESS_NONE) {
6886 ret = uncompress_inline(path, inode, page,
6887 pg_offset,
6888 extent_offset, item);
6889 if (ret) {
6890 err = ret;
6891 goto out;
6892 }
6893 } else {
6894 map = kmap(page);
6895 read_extent_buffer(leaf, map + pg_offset, ptr,
6896 copy_size);
6897 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6898 memset(map + pg_offset + copy_size, 0,
6899 PAGE_CACHE_SIZE - pg_offset -
6900 copy_size);
6901 }
6902 kunmap(page);
6903 }
6904 flush_dcache_page(page);
6905 } else if (create && PageUptodate(page)) {
6906 BUG();
6907 if (!trans) {
6908 kunmap(page);
6909 free_extent_map(em);
6910 em = NULL;
6911
6912 btrfs_release_path(path);
6913 trans = btrfs_join_transaction(root);
6914
6915 if (IS_ERR(trans))
6916 return ERR_CAST(trans);
6917 goto again;
6918 }
6919 map = kmap(page);
6920 write_extent_buffer(leaf, map + pg_offset, ptr,
6921 copy_size);
6922 kunmap(page);
6923 btrfs_mark_buffer_dirty(leaf);
6924 }
6925 set_extent_uptodate(io_tree, em->start,
6926 extent_map_end(em) - 1, NULL, GFP_NOFS);
6927 goto insert;
6928 }
6929 not_found:
6930 em->start = start;
6931 em->orig_start = start;
6932 em->len = len;
6933 not_found_em:
6934 em->block_start = EXTENT_MAP_HOLE;
6935 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6936 insert:
6937 btrfs_release_path(path);
6938 if (em->start > start || extent_map_end(em) <= start) {
6939 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6940 em->start, em->len, start, len);
6941 err = -EIO;
6942 goto out;
6943 }
6944
6945 err = 0;
6946 write_lock(&em_tree->lock);
6947 ret = add_extent_mapping(em_tree, em, 0);
6948 /* it is possible that someone inserted the extent into the tree
6949 * while we had the lock dropped. It is also possible that
6950 * an overlapping map exists in the tree
6951 */
6952 if (ret == -EEXIST) {
6953 struct extent_map *existing;
6954
6955 ret = 0;
6956
6957 existing = search_extent_mapping(em_tree, start, len);
6958 /*
6959 * existing will always be non-NULL, since there must be
6960 * extent causing the -EEXIST.
6961 */
6962 if (start >= extent_map_end(existing) ||
6963 start <= existing->start) {
6964 /*
6965 * The existing extent map is the one nearest to
6966 * the [start, start + len) range which overlaps
6967 */
6968 err = merge_extent_mapping(em_tree, existing,
6969 em, start);
6970 free_extent_map(existing);
6971 if (err) {
6972 free_extent_map(em);
6973 em = NULL;
6974 }
6975 } else {
6976 free_extent_map(em);
6977 em = existing;
6978 err = 0;
6979 }
6980 }
6981 write_unlock(&em_tree->lock);
6982 out:
6983
6984 trace_btrfs_get_extent(root, em);
6985
6986 btrfs_free_path(path);
6987 if (trans) {
6988 ret = btrfs_end_transaction(trans, root);
6989 if (!err)
6990 err = ret;
6991 }
6992 if (err) {
6993 free_extent_map(em);
6994 return ERR_PTR(err);
6995 }
6996 BUG_ON(!em); /* Error is always set */
6997 return em;
6998 }
6999
7000 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7001 size_t pg_offset, u64 start, u64 len,
7002 int create)
7003 {
7004 struct extent_map *em;
7005 struct extent_map *hole_em = NULL;
7006 u64 range_start = start;
7007 u64 end;
7008 u64 found;
7009 u64 found_end;
7010 int err = 0;
7011
7012 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7013 if (IS_ERR(em))
7014 return em;
7015 if (em) {
7016 /*
7017 * if our em maps to
7018 * - a hole or
7019 * - a pre-alloc extent,
7020 * there might actually be delalloc bytes behind it.
7021 */
7022 if (em->block_start != EXTENT_MAP_HOLE &&
7023 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7024 return em;
7025 else
7026 hole_em = em;
7027 }
7028
7029 /* check to see if we've wrapped (len == -1 or similar) */
7030 end = start + len;
7031 if (end < start)
7032 end = (u64)-1;
7033 else
7034 end -= 1;
7035
7036 em = NULL;
7037
7038 /* ok, we didn't find anything, lets look for delalloc */
7039 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7040 end, len, EXTENT_DELALLOC, 1);
7041 found_end = range_start + found;
7042 if (found_end < range_start)
7043 found_end = (u64)-1;
7044
7045 /*
7046 * we didn't find anything useful, return
7047 * the original results from get_extent()
7048 */
7049 if (range_start > end || found_end <= start) {
7050 em = hole_em;
7051 hole_em = NULL;
7052 goto out;
7053 }
7054
7055 /* adjust the range_start to make sure it doesn't
7056 * go backwards from the start they passed in
7057 */
7058 range_start = max(start, range_start);
7059 found = found_end - range_start;
7060
7061 if (found > 0) {
7062 u64 hole_start = start;
7063 u64 hole_len = len;
7064
7065 em = alloc_extent_map();
7066 if (!em) {
7067 err = -ENOMEM;
7068 goto out;
7069 }
7070 /*
7071 * when btrfs_get_extent can't find anything it
7072 * returns one huge hole
7073 *
7074 * make sure what it found really fits our range, and
7075 * adjust to make sure it is based on the start from
7076 * the caller
7077 */
7078 if (hole_em) {
7079 u64 calc_end = extent_map_end(hole_em);
7080
7081 if (calc_end <= start || (hole_em->start > end)) {
7082 free_extent_map(hole_em);
7083 hole_em = NULL;
7084 } else {
7085 hole_start = max(hole_em->start, start);
7086 hole_len = calc_end - hole_start;
7087 }
7088 }
7089 em->bdev = NULL;
7090 if (hole_em && range_start > hole_start) {
7091 /* our hole starts before our delalloc, so we
7092 * have to return just the parts of the hole
7093 * that go until the delalloc starts
7094 */
7095 em->len = min(hole_len,
7096 range_start - hole_start);
7097 em->start = hole_start;
7098 em->orig_start = hole_start;
7099 /*
7100 * don't adjust block start at all,
7101 * it is fixed at EXTENT_MAP_HOLE
7102 */
7103 em->block_start = hole_em->block_start;
7104 em->block_len = hole_len;
7105 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7106 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7107 } else {
7108 em->start = range_start;
7109 em->len = found;
7110 em->orig_start = range_start;
7111 em->block_start = EXTENT_MAP_DELALLOC;
7112 em->block_len = found;
7113 }
7114 } else if (hole_em) {
7115 return hole_em;
7116 }
7117 out:
7118
7119 free_extent_map(hole_em);
7120 if (err) {
7121 free_extent_map(em);
7122 return ERR_PTR(err);
7123 }
7124 return em;
7125 }
7126
7127 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7128 u64 start, u64 len)
7129 {
7130 struct btrfs_root *root = BTRFS_I(inode)->root;
7131 struct extent_map *em;
7132 struct btrfs_key ins;
7133 u64 alloc_hint;
7134 int ret;
7135
7136 alloc_hint = get_extent_allocation_hint(inode, start, len);
7137 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7138 alloc_hint, &ins, 1, 1);
7139 if (ret)
7140 return ERR_PTR(ret);
7141
7142 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7143 ins.offset, ins.offset, ins.offset, 0);
7144 if (IS_ERR(em)) {
7145 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7146 return em;
7147 }
7148
7149 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7150 ins.offset, ins.offset, 0);
7151 if (ret) {
7152 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7153 free_extent_map(em);
7154 return ERR_PTR(ret);
7155 }
7156
7157 return em;
7158 }
7159
7160 /*
7161 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7162 * block must be cow'd
7163 */
7164 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7165 u64 *orig_start, u64 *orig_block_len,
7166 u64 *ram_bytes)
7167 {
7168 struct btrfs_trans_handle *trans;
7169 struct btrfs_path *path;
7170 int ret;
7171 struct extent_buffer *leaf;
7172 struct btrfs_root *root = BTRFS_I(inode)->root;
7173 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7174 struct btrfs_file_extent_item *fi;
7175 struct btrfs_key key;
7176 u64 disk_bytenr;
7177 u64 backref_offset;
7178 u64 extent_end;
7179 u64 num_bytes;
7180 int slot;
7181 int found_type;
7182 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7183
7184 path = btrfs_alloc_path();
7185 if (!path)
7186 return -ENOMEM;
7187
7188 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7189 offset, 0);
7190 if (ret < 0)
7191 goto out;
7192
7193 slot = path->slots[0];
7194 if (ret == 1) {
7195 if (slot == 0) {
7196 /* can't find the item, must cow */
7197 ret = 0;
7198 goto out;
7199 }
7200 slot--;
7201 }
7202 ret = 0;
7203 leaf = path->nodes[0];
7204 btrfs_item_key_to_cpu(leaf, &key, slot);
7205 if (key.objectid != btrfs_ino(inode) ||
7206 key.type != BTRFS_EXTENT_DATA_KEY) {
7207 /* not our file or wrong item type, must cow */
7208 goto out;
7209 }
7210
7211 if (key.offset > offset) {
7212 /* Wrong offset, must cow */
7213 goto out;
7214 }
7215
7216 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7217 found_type = btrfs_file_extent_type(leaf, fi);
7218 if (found_type != BTRFS_FILE_EXTENT_REG &&
7219 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7220 /* not a regular extent, must cow */
7221 goto out;
7222 }
7223
7224 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7225 goto out;
7226
7227 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7228 if (extent_end <= offset)
7229 goto out;
7230
7231 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7232 if (disk_bytenr == 0)
7233 goto out;
7234
7235 if (btrfs_file_extent_compression(leaf, fi) ||
7236 btrfs_file_extent_encryption(leaf, fi) ||
7237 btrfs_file_extent_other_encoding(leaf, fi))
7238 goto out;
7239
7240 backref_offset = btrfs_file_extent_offset(leaf, fi);
7241
7242 if (orig_start) {
7243 *orig_start = key.offset - backref_offset;
7244 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7245 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7246 }
7247
7248 if (btrfs_extent_readonly(root, disk_bytenr))
7249 goto out;
7250
7251 num_bytes = min(offset + *len, extent_end) - offset;
7252 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7253 u64 range_end;
7254
7255 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7256 ret = test_range_bit(io_tree, offset, range_end,
7257 EXTENT_DELALLOC, 0, NULL);
7258 if (ret) {
7259 ret = -EAGAIN;
7260 goto out;
7261 }
7262 }
7263
7264 btrfs_release_path(path);
7265
7266 /*
7267 * look for other files referencing this extent, if we
7268 * find any we must cow
7269 */
7270 trans = btrfs_join_transaction(root);
7271 if (IS_ERR(trans)) {
7272 ret = 0;
7273 goto out;
7274 }
7275
7276 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7277 key.offset - backref_offset, disk_bytenr);
7278 btrfs_end_transaction(trans, root);
7279 if (ret) {
7280 ret = 0;
7281 goto out;
7282 }
7283
7284 /*
7285 * adjust disk_bytenr and num_bytes to cover just the bytes
7286 * in this extent we are about to write. If there
7287 * are any csums in that range we have to cow in order
7288 * to keep the csums correct
7289 */
7290 disk_bytenr += backref_offset;
7291 disk_bytenr += offset - key.offset;
7292 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7293 goto out;
7294 /*
7295 * all of the above have passed, it is safe to overwrite this extent
7296 * without cow
7297 */
7298 *len = num_bytes;
7299 ret = 1;
7300 out:
7301 btrfs_free_path(path);
7302 return ret;
7303 }
7304
7305 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7306 {
7307 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7308 int found = false;
7309 void **pagep = NULL;
7310 struct page *page = NULL;
7311 int start_idx;
7312 int end_idx;
7313
7314 start_idx = start >> PAGE_CACHE_SHIFT;
7315
7316 /*
7317 * end is the last byte in the last page. end == start is legal
7318 */
7319 end_idx = end >> PAGE_CACHE_SHIFT;
7320
7321 rcu_read_lock();
7322
7323 /* Most of the code in this while loop is lifted from
7324 * find_get_page. It's been modified to begin searching from a
7325 * page and return just the first page found in that range. If the
7326 * found idx is less than or equal to the end idx then we know that
7327 * a page exists. If no pages are found or if those pages are
7328 * outside of the range then we're fine (yay!) */
7329 while (page == NULL &&
7330 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7331 page = radix_tree_deref_slot(pagep);
7332 if (unlikely(!page))
7333 break;
7334
7335 if (radix_tree_exception(page)) {
7336 if (radix_tree_deref_retry(page)) {
7337 page = NULL;
7338 continue;
7339 }
7340 /*
7341 * Otherwise, shmem/tmpfs must be storing a swap entry
7342 * here as an exceptional entry: so return it without
7343 * attempting to raise page count.
7344 */
7345 page = NULL;
7346 break; /* TODO: Is this relevant for this use case? */
7347 }
7348
7349 if (!page_cache_get_speculative(page)) {
7350 page = NULL;
7351 continue;
7352 }
7353
7354 /*
7355 * Has the page moved?
7356 * This is part of the lockless pagecache protocol. See
7357 * include/linux/pagemap.h for details.
7358 */
7359 if (unlikely(page != *pagep)) {
7360 page_cache_release(page);
7361 page = NULL;
7362 }
7363 }
7364
7365 if (page) {
7366 if (page->index <= end_idx)
7367 found = true;
7368 page_cache_release(page);
7369 }
7370
7371 rcu_read_unlock();
7372 return found;
7373 }
7374
7375 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7376 struct extent_state **cached_state, int writing)
7377 {
7378 struct btrfs_ordered_extent *ordered;
7379 int ret = 0;
7380
7381 while (1) {
7382 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7383 0, cached_state);
7384 /*
7385 * We're concerned with the entire range that we're going to be
7386 * doing DIO to, so we need to make sure theres no ordered
7387 * extents in this range.
7388 */
7389 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7390 lockend - lockstart + 1);
7391
7392 /*
7393 * We need to make sure there are no buffered pages in this
7394 * range either, we could have raced between the invalidate in
7395 * generic_file_direct_write and locking the extent. The
7396 * invalidate needs to happen so that reads after a write do not
7397 * get stale data.
7398 */
7399 if (!ordered &&
7400 (!writing ||
7401 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7402 break;
7403
7404 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7405 cached_state, GFP_NOFS);
7406
7407 if (ordered) {
7408 btrfs_start_ordered_extent(inode, ordered, 1);
7409 btrfs_put_ordered_extent(ordered);
7410 } else {
7411 /* Screw you mmap */
7412 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7413 if (ret)
7414 break;
7415 ret = filemap_fdatawait_range(inode->i_mapping,
7416 lockstart,
7417 lockend);
7418 if (ret)
7419 break;
7420
7421 /*
7422 * If we found a page that couldn't be invalidated just
7423 * fall back to buffered.
7424 */
7425 ret = invalidate_inode_pages2_range(inode->i_mapping,
7426 lockstart >> PAGE_CACHE_SHIFT,
7427 lockend >> PAGE_CACHE_SHIFT);
7428 if (ret)
7429 break;
7430 }
7431
7432 cond_resched();
7433 }
7434
7435 return ret;
7436 }
7437
7438 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7439 u64 len, u64 orig_start,
7440 u64 block_start, u64 block_len,
7441 u64 orig_block_len, u64 ram_bytes,
7442 int type)
7443 {
7444 struct extent_map_tree *em_tree;
7445 struct extent_map *em;
7446 struct btrfs_root *root = BTRFS_I(inode)->root;
7447 int ret;
7448
7449 em_tree = &BTRFS_I(inode)->extent_tree;
7450 em = alloc_extent_map();
7451 if (!em)
7452 return ERR_PTR(-ENOMEM);
7453
7454 em->start = start;
7455 em->orig_start = orig_start;
7456 em->mod_start = start;
7457 em->mod_len = len;
7458 em->len = len;
7459 em->block_len = block_len;
7460 em->block_start = block_start;
7461 em->bdev = root->fs_info->fs_devices->latest_bdev;
7462 em->orig_block_len = orig_block_len;
7463 em->ram_bytes = ram_bytes;
7464 em->generation = -1;
7465 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7466 if (type == BTRFS_ORDERED_PREALLOC)
7467 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7468
7469 do {
7470 btrfs_drop_extent_cache(inode, em->start,
7471 em->start + em->len - 1, 0);
7472 write_lock(&em_tree->lock);
7473 ret = add_extent_mapping(em_tree, em, 1);
7474 write_unlock(&em_tree->lock);
7475 } while (ret == -EEXIST);
7476
7477 if (ret) {
7478 free_extent_map(em);
7479 return ERR_PTR(ret);
7480 }
7481
7482 return em;
7483 }
7484
7485 struct btrfs_dio_data {
7486 u64 outstanding_extents;
7487 u64 reserve;
7488 };
7489
7490 static void adjust_dio_outstanding_extents(struct inode *inode,
7491 struct btrfs_dio_data *dio_data,
7492 const u64 len)
7493 {
7494 unsigned num_extents;
7495
7496 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7497 BTRFS_MAX_EXTENT_SIZE);
7498 /*
7499 * If we have an outstanding_extents count still set then we're
7500 * within our reservation, otherwise we need to adjust our inode
7501 * counter appropriately.
7502 */
7503 if (dio_data->outstanding_extents) {
7504 dio_data->outstanding_extents -= num_extents;
7505 } else {
7506 spin_lock(&BTRFS_I(inode)->lock);
7507 BTRFS_I(inode)->outstanding_extents += num_extents;
7508 spin_unlock(&BTRFS_I(inode)->lock);
7509 }
7510 }
7511
7512 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7513 struct buffer_head *bh_result, int create)
7514 {
7515 struct extent_map *em;
7516 struct btrfs_root *root = BTRFS_I(inode)->root;
7517 struct extent_state *cached_state = NULL;
7518 struct btrfs_dio_data *dio_data = NULL;
7519 u64 start = iblock << inode->i_blkbits;
7520 u64 lockstart, lockend;
7521 u64 len = bh_result->b_size;
7522 int unlock_bits = EXTENT_LOCKED;
7523 int ret = 0;
7524
7525 if (create)
7526 unlock_bits |= EXTENT_DIRTY;
7527 else
7528 len = min_t(u64, len, root->sectorsize);
7529
7530 lockstart = start;
7531 lockend = start + len - 1;
7532
7533 if (current->journal_info) {
7534 /*
7535 * Need to pull our outstanding extents and set journal_info to NULL so
7536 * that anything that needs to check if there's a transction doesn't get
7537 * confused.
7538 */
7539 dio_data = current->journal_info;
7540 current->journal_info = NULL;
7541 }
7542
7543 /*
7544 * If this errors out it's because we couldn't invalidate pagecache for
7545 * this range and we need to fallback to buffered.
7546 */
7547 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7548 create)) {
7549 ret = -ENOTBLK;
7550 goto err;
7551 }
7552
7553 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7554 if (IS_ERR(em)) {
7555 ret = PTR_ERR(em);
7556 goto unlock_err;
7557 }
7558
7559 /*
7560 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7561 * io. INLINE is special, and we could probably kludge it in here, but
7562 * it's still buffered so for safety lets just fall back to the generic
7563 * buffered path.
7564 *
7565 * For COMPRESSED we _have_ to read the entire extent in so we can
7566 * decompress it, so there will be buffering required no matter what we
7567 * do, so go ahead and fallback to buffered.
7568 *
7569 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7570 * to buffered IO. Don't blame me, this is the price we pay for using
7571 * the generic code.
7572 */
7573 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7574 em->block_start == EXTENT_MAP_INLINE) {
7575 free_extent_map(em);
7576 ret = -ENOTBLK;
7577 goto unlock_err;
7578 }
7579
7580 /* Just a good old fashioned hole, return */
7581 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7582 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7583 free_extent_map(em);
7584 goto unlock_err;
7585 }
7586
7587 /*
7588 * We don't allocate a new extent in the following cases
7589 *
7590 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7591 * existing extent.
7592 * 2) The extent is marked as PREALLOC. We're good to go here and can
7593 * just use the extent.
7594 *
7595 */
7596 if (!create) {
7597 len = min(len, em->len - (start - em->start));
7598 lockstart = start + len;
7599 goto unlock;
7600 }
7601
7602 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7603 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7604 em->block_start != EXTENT_MAP_HOLE)) {
7605 int type;
7606 u64 block_start, orig_start, orig_block_len, ram_bytes;
7607
7608 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7609 type = BTRFS_ORDERED_PREALLOC;
7610 else
7611 type = BTRFS_ORDERED_NOCOW;
7612 len = min(len, em->len - (start - em->start));
7613 block_start = em->block_start + (start - em->start);
7614
7615 if (can_nocow_extent(inode, start, &len, &orig_start,
7616 &orig_block_len, &ram_bytes) == 1) {
7617 if (type == BTRFS_ORDERED_PREALLOC) {
7618 free_extent_map(em);
7619 em = create_pinned_em(inode, start, len,
7620 orig_start,
7621 block_start, len,
7622 orig_block_len,
7623 ram_bytes, type);
7624 if (IS_ERR(em)) {
7625 ret = PTR_ERR(em);
7626 goto unlock_err;
7627 }
7628 }
7629
7630 ret = btrfs_add_ordered_extent_dio(inode, start,
7631 block_start, len, len, type);
7632 if (ret) {
7633 free_extent_map(em);
7634 goto unlock_err;
7635 }
7636 goto unlock;
7637 }
7638 }
7639
7640 /*
7641 * this will cow the extent, reset the len in case we changed
7642 * it above
7643 */
7644 len = bh_result->b_size;
7645 free_extent_map(em);
7646 em = btrfs_new_extent_direct(inode, start, len);
7647 if (IS_ERR(em)) {
7648 ret = PTR_ERR(em);
7649 goto unlock_err;
7650 }
7651 len = min(len, em->len - (start - em->start));
7652 unlock:
7653 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7654 inode->i_blkbits;
7655 bh_result->b_size = len;
7656 bh_result->b_bdev = em->bdev;
7657 set_buffer_mapped(bh_result);
7658 if (create) {
7659 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7660 set_buffer_new(bh_result);
7661
7662 /*
7663 * Need to update the i_size under the extent lock so buffered
7664 * readers will get the updated i_size when we unlock.
7665 */
7666 if (start + len > i_size_read(inode))
7667 i_size_write(inode, start + len);
7668
7669 adjust_dio_outstanding_extents(inode, dio_data, len);
7670 btrfs_free_reserved_data_space(inode, start, len);
7671 WARN_ON(dio_data->reserve < len);
7672 dio_data->reserve -= len;
7673 current->journal_info = dio_data;
7674 }
7675
7676 /*
7677 * In the case of write we need to clear and unlock the entire range,
7678 * in the case of read we need to unlock only the end area that we
7679 * aren't using if there is any left over space.
7680 */
7681 if (lockstart < lockend) {
7682 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7683 lockend, unlock_bits, 1, 0,
7684 &cached_state, GFP_NOFS);
7685 } else {
7686 free_extent_state(cached_state);
7687 }
7688
7689 free_extent_map(em);
7690
7691 return 0;
7692
7693 unlock_err:
7694 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7695 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7696 err:
7697 if (dio_data)
7698 current->journal_info = dio_data;
7699 /*
7700 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7701 * write less data then expected, so that we don't underflow our inode's
7702 * outstanding extents counter.
7703 */
7704 if (create && dio_data)
7705 adjust_dio_outstanding_extents(inode, dio_data, len);
7706
7707 return ret;
7708 }
7709
7710 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7711 int rw, int mirror_num)
7712 {
7713 struct btrfs_root *root = BTRFS_I(inode)->root;
7714 int ret;
7715
7716 BUG_ON(rw & REQ_WRITE);
7717
7718 bio_get(bio);
7719
7720 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7721 BTRFS_WQ_ENDIO_DIO_REPAIR);
7722 if (ret)
7723 goto err;
7724
7725 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7726 err:
7727 bio_put(bio);
7728 return ret;
7729 }
7730
7731 static int btrfs_check_dio_repairable(struct inode *inode,
7732 struct bio *failed_bio,
7733 struct io_failure_record *failrec,
7734 int failed_mirror)
7735 {
7736 int num_copies;
7737
7738 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7739 failrec->logical, failrec->len);
7740 if (num_copies == 1) {
7741 /*
7742 * we only have a single copy of the data, so don't bother with
7743 * all the retry and error correction code that follows. no
7744 * matter what the error is, it is very likely to persist.
7745 */
7746 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7747 num_copies, failrec->this_mirror, failed_mirror);
7748 return 0;
7749 }
7750
7751 failrec->failed_mirror = failed_mirror;
7752 failrec->this_mirror++;
7753 if (failrec->this_mirror == failed_mirror)
7754 failrec->this_mirror++;
7755
7756 if (failrec->this_mirror > num_copies) {
7757 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7758 num_copies, failrec->this_mirror, failed_mirror);
7759 return 0;
7760 }
7761
7762 return 1;
7763 }
7764
7765 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7766 struct page *page, u64 start, u64 end,
7767 int failed_mirror, bio_end_io_t *repair_endio,
7768 void *repair_arg)
7769 {
7770 struct io_failure_record *failrec;
7771 struct bio *bio;
7772 int isector;
7773 int read_mode;
7774 int ret;
7775
7776 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7777
7778 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7779 if (ret)
7780 return ret;
7781
7782 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7783 failed_mirror);
7784 if (!ret) {
7785 free_io_failure(inode, failrec);
7786 return -EIO;
7787 }
7788
7789 if (failed_bio->bi_vcnt > 1)
7790 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7791 else
7792 read_mode = READ_SYNC;
7793
7794 isector = start - btrfs_io_bio(failed_bio)->logical;
7795 isector >>= inode->i_sb->s_blocksize_bits;
7796 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7797 0, isector, repair_endio, repair_arg);
7798 if (!bio) {
7799 free_io_failure(inode, failrec);
7800 return -EIO;
7801 }
7802
7803 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7804 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7805 read_mode, failrec->this_mirror, failrec->in_validation);
7806
7807 ret = submit_dio_repair_bio(inode, bio, read_mode,
7808 failrec->this_mirror);
7809 if (ret) {
7810 free_io_failure(inode, failrec);
7811 bio_put(bio);
7812 }
7813
7814 return ret;
7815 }
7816
7817 struct btrfs_retry_complete {
7818 struct completion done;
7819 struct inode *inode;
7820 u64 start;
7821 int uptodate;
7822 };
7823
7824 static void btrfs_retry_endio_nocsum(struct bio *bio)
7825 {
7826 struct btrfs_retry_complete *done = bio->bi_private;
7827 struct bio_vec *bvec;
7828 int i;
7829
7830 if (bio->bi_error)
7831 goto end;
7832
7833 done->uptodate = 1;
7834 bio_for_each_segment_all(bvec, bio, i)
7835 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7836 end:
7837 complete(&done->done);
7838 bio_put(bio);
7839 }
7840
7841 static int __btrfs_correct_data_nocsum(struct inode *inode,
7842 struct btrfs_io_bio *io_bio)
7843 {
7844 struct bio_vec *bvec;
7845 struct btrfs_retry_complete done;
7846 u64 start;
7847 int i;
7848 int ret;
7849
7850 start = io_bio->logical;
7851 done.inode = inode;
7852
7853 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7854 try_again:
7855 done.uptodate = 0;
7856 done.start = start;
7857 init_completion(&done.done);
7858
7859 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7860 start + bvec->bv_len - 1,
7861 io_bio->mirror_num,
7862 btrfs_retry_endio_nocsum, &done);
7863 if (ret)
7864 return ret;
7865
7866 wait_for_completion(&done.done);
7867
7868 if (!done.uptodate) {
7869 /* We might have another mirror, so try again */
7870 goto try_again;
7871 }
7872
7873 start += bvec->bv_len;
7874 }
7875
7876 return 0;
7877 }
7878
7879 static void btrfs_retry_endio(struct bio *bio)
7880 {
7881 struct btrfs_retry_complete *done = bio->bi_private;
7882 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7883 struct bio_vec *bvec;
7884 int uptodate;
7885 int ret;
7886 int i;
7887
7888 if (bio->bi_error)
7889 goto end;
7890
7891 uptodate = 1;
7892 bio_for_each_segment_all(bvec, bio, i) {
7893 ret = __readpage_endio_check(done->inode, io_bio, i,
7894 bvec->bv_page, 0,
7895 done->start, bvec->bv_len);
7896 if (!ret)
7897 clean_io_failure(done->inode, done->start,
7898 bvec->bv_page, 0);
7899 else
7900 uptodate = 0;
7901 }
7902
7903 done->uptodate = uptodate;
7904 end:
7905 complete(&done->done);
7906 bio_put(bio);
7907 }
7908
7909 static int __btrfs_subio_endio_read(struct inode *inode,
7910 struct btrfs_io_bio *io_bio, int err)
7911 {
7912 struct bio_vec *bvec;
7913 struct btrfs_retry_complete done;
7914 u64 start;
7915 u64 offset = 0;
7916 int i;
7917 int ret;
7918
7919 err = 0;
7920 start = io_bio->logical;
7921 done.inode = inode;
7922
7923 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7924 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7925 0, start, bvec->bv_len);
7926 if (likely(!ret))
7927 goto next;
7928 try_again:
7929 done.uptodate = 0;
7930 done.start = start;
7931 init_completion(&done.done);
7932
7933 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7934 start + bvec->bv_len - 1,
7935 io_bio->mirror_num,
7936 btrfs_retry_endio, &done);
7937 if (ret) {
7938 err = ret;
7939 goto next;
7940 }
7941
7942 wait_for_completion(&done.done);
7943
7944 if (!done.uptodate) {
7945 /* We might have another mirror, so try again */
7946 goto try_again;
7947 }
7948 next:
7949 offset += bvec->bv_len;
7950 start += bvec->bv_len;
7951 }
7952
7953 return err;
7954 }
7955
7956 static int btrfs_subio_endio_read(struct inode *inode,
7957 struct btrfs_io_bio *io_bio, int err)
7958 {
7959 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7960
7961 if (skip_csum) {
7962 if (unlikely(err))
7963 return __btrfs_correct_data_nocsum(inode, io_bio);
7964 else
7965 return 0;
7966 } else {
7967 return __btrfs_subio_endio_read(inode, io_bio, err);
7968 }
7969 }
7970
7971 static void btrfs_endio_direct_read(struct bio *bio)
7972 {
7973 struct btrfs_dio_private *dip = bio->bi_private;
7974 struct inode *inode = dip->inode;
7975 struct bio *dio_bio;
7976 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7977 int err = bio->bi_error;
7978
7979 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7980 err = btrfs_subio_endio_read(inode, io_bio, err);
7981
7982 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7983 dip->logical_offset + dip->bytes - 1);
7984 dio_bio = dip->dio_bio;
7985
7986 kfree(dip);
7987
7988 dio_end_io(dio_bio, bio->bi_error);
7989
7990 if (io_bio->end_io)
7991 io_bio->end_io(io_bio, err);
7992 bio_put(bio);
7993 }
7994
7995 static void btrfs_endio_direct_write(struct bio *bio)
7996 {
7997 struct btrfs_dio_private *dip = bio->bi_private;
7998 struct inode *inode = dip->inode;
7999 struct btrfs_root *root = BTRFS_I(inode)->root;
8000 struct btrfs_ordered_extent *ordered = NULL;
8001 u64 ordered_offset = dip->logical_offset;
8002 u64 ordered_bytes = dip->bytes;
8003 struct bio *dio_bio;
8004 int ret;
8005
8006 again:
8007 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8008 &ordered_offset,
8009 ordered_bytes,
8010 !bio->bi_error);
8011 if (!ret)
8012 goto out_test;
8013
8014 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8015 finish_ordered_fn, NULL, NULL);
8016 btrfs_queue_work(root->fs_info->endio_write_workers,
8017 &ordered->work);
8018 out_test:
8019 /*
8020 * our bio might span multiple ordered extents. If we haven't
8021 * completed the accounting for the whole dio, go back and try again
8022 */
8023 if (ordered_offset < dip->logical_offset + dip->bytes) {
8024 ordered_bytes = dip->logical_offset + dip->bytes -
8025 ordered_offset;
8026 ordered = NULL;
8027 goto again;
8028 }
8029 dio_bio = dip->dio_bio;
8030
8031 kfree(dip);
8032
8033 dio_end_io(dio_bio, bio->bi_error);
8034 bio_put(bio);
8035 }
8036
8037 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8038 struct bio *bio, int mirror_num,
8039 unsigned long bio_flags, u64 offset)
8040 {
8041 int ret;
8042 struct btrfs_root *root = BTRFS_I(inode)->root;
8043 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8044 BUG_ON(ret); /* -ENOMEM */
8045 return 0;
8046 }
8047
8048 static void btrfs_end_dio_bio(struct bio *bio)
8049 {
8050 struct btrfs_dio_private *dip = bio->bi_private;
8051 int err = bio->bi_error;
8052
8053 if (err)
8054 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8055 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8056 btrfs_ino(dip->inode), bio->bi_rw,
8057 (unsigned long long)bio->bi_iter.bi_sector,
8058 bio->bi_iter.bi_size, err);
8059
8060 if (dip->subio_endio)
8061 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8062
8063 if (err) {
8064 dip->errors = 1;
8065
8066 /*
8067 * before atomic variable goto zero, we must make sure
8068 * dip->errors is perceived to be set.
8069 */
8070 smp_mb__before_atomic();
8071 }
8072
8073 /* if there are more bios still pending for this dio, just exit */
8074 if (!atomic_dec_and_test(&dip->pending_bios))
8075 goto out;
8076
8077 if (dip->errors) {
8078 bio_io_error(dip->orig_bio);
8079 } else {
8080 dip->dio_bio->bi_error = 0;
8081 bio_endio(dip->orig_bio);
8082 }
8083 out:
8084 bio_put(bio);
8085 }
8086
8087 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8088 u64 first_sector, gfp_t gfp_flags)
8089 {
8090 struct bio *bio;
8091 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8092 if (bio)
8093 bio_associate_current(bio);
8094 return bio;
8095 }
8096
8097 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8098 struct inode *inode,
8099 struct btrfs_dio_private *dip,
8100 struct bio *bio,
8101 u64 file_offset)
8102 {
8103 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8104 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8105 int ret;
8106
8107 /*
8108 * We load all the csum data we need when we submit
8109 * the first bio to reduce the csum tree search and
8110 * contention.
8111 */
8112 if (dip->logical_offset == file_offset) {
8113 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8114 file_offset);
8115 if (ret)
8116 return ret;
8117 }
8118
8119 if (bio == dip->orig_bio)
8120 return 0;
8121
8122 file_offset -= dip->logical_offset;
8123 file_offset >>= inode->i_sb->s_blocksize_bits;
8124 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8125
8126 return 0;
8127 }
8128
8129 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8130 int rw, u64 file_offset, int skip_sum,
8131 int async_submit)
8132 {
8133 struct btrfs_dio_private *dip = bio->bi_private;
8134 int write = rw & REQ_WRITE;
8135 struct btrfs_root *root = BTRFS_I(inode)->root;
8136 int ret;
8137
8138 if (async_submit)
8139 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8140
8141 bio_get(bio);
8142
8143 if (!write) {
8144 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8145 BTRFS_WQ_ENDIO_DATA);
8146 if (ret)
8147 goto err;
8148 }
8149
8150 if (skip_sum)
8151 goto map;
8152
8153 if (write && async_submit) {
8154 ret = btrfs_wq_submit_bio(root->fs_info,
8155 inode, rw, bio, 0, 0,
8156 file_offset,
8157 __btrfs_submit_bio_start_direct_io,
8158 __btrfs_submit_bio_done);
8159 goto err;
8160 } else if (write) {
8161 /*
8162 * If we aren't doing async submit, calculate the csum of the
8163 * bio now.
8164 */
8165 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8166 if (ret)
8167 goto err;
8168 } else {
8169 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8170 file_offset);
8171 if (ret)
8172 goto err;
8173 }
8174 map:
8175 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8176 err:
8177 bio_put(bio);
8178 return ret;
8179 }
8180
8181 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8182 int skip_sum)
8183 {
8184 struct inode *inode = dip->inode;
8185 struct btrfs_root *root = BTRFS_I(inode)->root;
8186 struct bio *bio;
8187 struct bio *orig_bio = dip->orig_bio;
8188 struct bio_vec *bvec = orig_bio->bi_io_vec;
8189 u64 start_sector = orig_bio->bi_iter.bi_sector;
8190 u64 file_offset = dip->logical_offset;
8191 u64 submit_len = 0;
8192 u64 map_length;
8193 int nr_pages = 0;
8194 int ret;
8195 int async_submit = 0;
8196
8197 map_length = orig_bio->bi_iter.bi_size;
8198 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8199 &map_length, NULL, 0);
8200 if (ret)
8201 return -EIO;
8202
8203 if (map_length >= orig_bio->bi_iter.bi_size) {
8204 bio = orig_bio;
8205 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8206 goto submit;
8207 }
8208
8209 /* async crcs make it difficult to collect full stripe writes. */
8210 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8211 async_submit = 0;
8212 else
8213 async_submit = 1;
8214
8215 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8216 if (!bio)
8217 return -ENOMEM;
8218
8219 bio->bi_private = dip;
8220 bio->bi_end_io = btrfs_end_dio_bio;
8221 btrfs_io_bio(bio)->logical = file_offset;
8222 atomic_inc(&dip->pending_bios);
8223
8224 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8225 if (map_length < submit_len + bvec->bv_len ||
8226 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8227 bvec->bv_offset) < bvec->bv_len) {
8228 /*
8229 * inc the count before we submit the bio so
8230 * we know the end IO handler won't happen before
8231 * we inc the count. Otherwise, the dip might get freed
8232 * before we're done setting it up
8233 */
8234 atomic_inc(&dip->pending_bios);
8235 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8236 file_offset, skip_sum,
8237 async_submit);
8238 if (ret) {
8239 bio_put(bio);
8240 atomic_dec(&dip->pending_bios);
8241 goto out_err;
8242 }
8243
8244 start_sector += submit_len >> 9;
8245 file_offset += submit_len;
8246
8247 submit_len = 0;
8248 nr_pages = 0;
8249
8250 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8251 start_sector, GFP_NOFS);
8252 if (!bio)
8253 goto out_err;
8254 bio->bi_private = dip;
8255 bio->bi_end_io = btrfs_end_dio_bio;
8256 btrfs_io_bio(bio)->logical = file_offset;
8257
8258 map_length = orig_bio->bi_iter.bi_size;
8259 ret = btrfs_map_block(root->fs_info, rw,
8260 start_sector << 9,
8261 &map_length, NULL, 0);
8262 if (ret) {
8263 bio_put(bio);
8264 goto out_err;
8265 }
8266 } else {
8267 submit_len += bvec->bv_len;
8268 nr_pages++;
8269 bvec++;
8270 }
8271 }
8272
8273 submit:
8274 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8275 async_submit);
8276 if (!ret)
8277 return 0;
8278
8279 bio_put(bio);
8280 out_err:
8281 dip->errors = 1;
8282 /*
8283 * before atomic variable goto zero, we must
8284 * make sure dip->errors is perceived to be set.
8285 */
8286 smp_mb__before_atomic();
8287 if (atomic_dec_and_test(&dip->pending_bios))
8288 bio_io_error(dip->orig_bio);
8289
8290 /* bio_end_io() will handle error, so we needn't return it */
8291 return 0;
8292 }
8293
8294 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8295 struct inode *inode, loff_t file_offset)
8296 {
8297 struct btrfs_dio_private *dip = NULL;
8298 struct bio *io_bio = NULL;
8299 struct btrfs_io_bio *btrfs_bio;
8300 int skip_sum;
8301 int write = rw & REQ_WRITE;
8302 int ret = 0;
8303
8304 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8305
8306 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8307 if (!io_bio) {
8308 ret = -ENOMEM;
8309 goto free_ordered;
8310 }
8311
8312 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8313 if (!dip) {
8314 ret = -ENOMEM;
8315 goto free_ordered;
8316 }
8317
8318 dip->private = dio_bio->bi_private;
8319 dip->inode = inode;
8320 dip->logical_offset = file_offset;
8321 dip->bytes = dio_bio->bi_iter.bi_size;
8322 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8323 io_bio->bi_private = dip;
8324 dip->orig_bio = io_bio;
8325 dip->dio_bio = dio_bio;
8326 atomic_set(&dip->pending_bios, 0);
8327 btrfs_bio = btrfs_io_bio(io_bio);
8328 btrfs_bio->logical = file_offset;
8329
8330 if (write) {
8331 io_bio->bi_end_io = btrfs_endio_direct_write;
8332 } else {
8333 io_bio->bi_end_io = btrfs_endio_direct_read;
8334 dip->subio_endio = btrfs_subio_endio_read;
8335 }
8336
8337 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8338 if (!ret)
8339 return;
8340
8341 if (btrfs_bio->end_io)
8342 btrfs_bio->end_io(btrfs_bio, ret);
8343
8344 free_ordered:
8345 /*
8346 * If we arrived here it means either we failed to submit the dip
8347 * or we either failed to clone the dio_bio or failed to allocate the
8348 * dip. If we cloned the dio_bio and allocated the dip, we can just
8349 * call bio_endio against our io_bio so that we get proper resource
8350 * cleanup if we fail to submit the dip, otherwise, we must do the
8351 * same as btrfs_endio_direct_[write|read] because we can't call these
8352 * callbacks - they require an allocated dip and a clone of dio_bio.
8353 */
8354 if (io_bio && dip) {
8355 io_bio->bi_error = -EIO;
8356 bio_endio(io_bio);
8357 /*
8358 * The end io callbacks free our dip, do the final put on io_bio
8359 * and all the cleanup and final put for dio_bio (through
8360 * dio_end_io()).
8361 */
8362 dip = NULL;
8363 io_bio = NULL;
8364 } else {
8365 if (write) {
8366 struct btrfs_ordered_extent *ordered;
8367
8368 ordered = btrfs_lookup_ordered_extent(inode,
8369 file_offset);
8370 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8371 /*
8372 * Decrements our ref on the ordered extent and removes
8373 * the ordered extent from the inode's ordered tree,
8374 * doing all the proper resource cleanup such as for the
8375 * reserved space and waking up any waiters for this
8376 * ordered extent (through btrfs_remove_ordered_extent).
8377 */
8378 btrfs_finish_ordered_io(ordered);
8379 } else {
8380 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8381 file_offset + dio_bio->bi_iter.bi_size - 1);
8382 }
8383 dio_bio->bi_error = -EIO;
8384 /*
8385 * Releases and cleans up our dio_bio, no need to bio_put()
8386 * nor bio_endio()/bio_io_error() against dio_bio.
8387 */
8388 dio_end_io(dio_bio, ret);
8389 }
8390 if (io_bio)
8391 bio_put(io_bio);
8392 kfree(dip);
8393 }
8394
8395 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8396 const struct iov_iter *iter, loff_t offset)
8397 {
8398 int seg;
8399 int i;
8400 unsigned blocksize_mask = root->sectorsize - 1;
8401 ssize_t retval = -EINVAL;
8402
8403 if (offset & blocksize_mask)
8404 goto out;
8405
8406 if (iov_iter_alignment(iter) & blocksize_mask)
8407 goto out;
8408
8409 /* If this is a write we don't need to check anymore */
8410 if (iov_iter_rw(iter) == WRITE)
8411 return 0;
8412 /*
8413 * Check to make sure we don't have duplicate iov_base's in this
8414 * iovec, if so return EINVAL, otherwise we'll get csum errors
8415 * when reading back.
8416 */
8417 for (seg = 0; seg < iter->nr_segs; seg++) {
8418 for (i = seg + 1; i < iter->nr_segs; i++) {
8419 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8420 goto out;
8421 }
8422 }
8423 retval = 0;
8424 out:
8425 return retval;
8426 }
8427
8428 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8429 loff_t offset)
8430 {
8431 struct file *file = iocb->ki_filp;
8432 struct inode *inode = file->f_mapping->host;
8433 struct btrfs_root *root = BTRFS_I(inode)->root;
8434 struct btrfs_dio_data dio_data = { 0 };
8435 size_t count = 0;
8436 int flags = 0;
8437 bool wakeup = true;
8438 bool relock = false;
8439 ssize_t ret;
8440
8441 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8442 return 0;
8443
8444 inode_dio_begin(inode);
8445 smp_mb__after_atomic();
8446
8447 /*
8448 * The generic stuff only does filemap_write_and_wait_range, which
8449 * isn't enough if we've written compressed pages to this area, so
8450 * we need to flush the dirty pages again to make absolutely sure
8451 * that any outstanding dirty pages are on disk.
8452 */
8453 count = iov_iter_count(iter);
8454 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8455 &BTRFS_I(inode)->runtime_flags))
8456 filemap_fdatawrite_range(inode->i_mapping, offset,
8457 offset + count - 1);
8458
8459 if (iov_iter_rw(iter) == WRITE) {
8460 /*
8461 * If the write DIO is beyond the EOF, we need update
8462 * the isize, but it is protected by i_mutex. So we can
8463 * not unlock the i_mutex at this case.
8464 */
8465 if (offset + count <= inode->i_size) {
8466 mutex_unlock(&inode->i_mutex);
8467 relock = true;
8468 }
8469 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8470 if (ret)
8471 goto out;
8472 dio_data.outstanding_extents = div64_u64(count +
8473 BTRFS_MAX_EXTENT_SIZE - 1,
8474 BTRFS_MAX_EXTENT_SIZE);
8475
8476 /*
8477 * We need to know how many extents we reserved so that we can
8478 * do the accounting properly if we go over the number we
8479 * originally calculated. Abuse current->journal_info for this.
8480 */
8481 dio_data.reserve = round_up(count, root->sectorsize);
8482 current->journal_info = &dio_data;
8483 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8484 &BTRFS_I(inode)->runtime_flags)) {
8485 inode_dio_end(inode);
8486 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8487 wakeup = false;
8488 }
8489
8490 ret = __blockdev_direct_IO(iocb, inode,
8491 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8492 iter, offset, btrfs_get_blocks_direct, NULL,
8493 btrfs_submit_direct, flags);
8494 if (iov_iter_rw(iter) == WRITE) {
8495 current->journal_info = NULL;
8496 if (ret < 0 && ret != -EIOCBQUEUED) {
8497 if (dio_data.reserve)
8498 btrfs_delalloc_release_space(inode, offset,
8499 dio_data.reserve);
8500 } else if (ret >= 0 && (size_t)ret < count)
8501 btrfs_delalloc_release_space(inode, offset,
8502 count - (size_t)ret);
8503 }
8504 out:
8505 if (wakeup)
8506 inode_dio_end(inode);
8507 if (relock)
8508 mutex_lock(&inode->i_mutex);
8509
8510 return ret;
8511 }
8512
8513 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8514
8515 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8516 __u64 start, __u64 len)
8517 {
8518 int ret;
8519
8520 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8521 if (ret)
8522 return ret;
8523
8524 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8525 }
8526
8527 int btrfs_readpage(struct file *file, struct page *page)
8528 {
8529 struct extent_io_tree *tree;
8530 tree = &BTRFS_I(page->mapping->host)->io_tree;
8531 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8532 }
8533
8534 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8535 {
8536 struct extent_io_tree *tree;
8537
8538
8539 if (current->flags & PF_MEMALLOC) {
8540 redirty_page_for_writepage(wbc, page);
8541 unlock_page(page);
8542 return 0;
8543 }
8544 tree = &BTRFS_I(page->mapping->host)->io_tree;
8545 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8546 }
8547
8548 static int btrfs_writepages(struct address_space *mapping,
8549 struct writeback_control *wbc)
8550 {
8551 struct extent_io_tree *tree;
8552
8553 tree = &BTRFS_I(mapping->host)->io_tree;
8554 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8555 }
8556
8557 static int
8558 btrfs_readpages(struct file *file, struct address_space *mapping,
8559 struct list_head *pages, unsigned nr_pages)
8560 {
8561 struct extent_io_tree *tree;
8562 tree = &BTRFS_I(mapping->host)->io_tree;
8563 return extent_readpages(tree, mapping, pages, nr_pages,
8564 btrfs_get_extent);
8565 }
8566 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8567 {
8568 struct extent_io_tree *tree;
8569 struct extent_map_tree *map;
8570 int ret;
8571
8572 tree = &BTRFS_I(page->mapping->host)->io_tree;
8573 map = &BTRFS_I(page->mapping->host)->extent_tree;
8574 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8575 if (ret == 1) {
8576 ClearPagePrivate(page);
8577 set_page_private(page, 0);
8578 page_cache_release(page);
8579 }
8580 return ret;
8581 }
8582
8583 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8584 {
8585 if (PageWriteback(page) || PageDirty(page))
8586 return 0;
8587 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8588 }
8589
8590 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8591 unsigned int length)
8592 {
8593 struct inode *inode = page->mapping->host;
8594 struct extent_io_tree *tree;
8595 struct btrfs_ordered_extent *ordered;
8596 struct extent_state *cached_state = NULL;
8597 u64 page_start = page_offset(page);
8598 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8599 int inode_evicting = inode->i_state & I_FREEING;
8600
8601 /*
8602 * we have the page locked, so new writeback can't start,
8603 * and the dirty bit won't be cleared while we are here.
8604 *
8605 * Wait for IO on this page so that we can safely clear
8606 * the PagePrivate2 bit and do ordered accounting
8607 */
8608 wait_on_page_writeback(page);
8609
8610 tree = &BTRFS_I(inode)->io_tree;
8611 if (offset) {
8612 btrfs_releasepage(page, GFP_NOFS);
8613 return;
8614 }
8615
8616 if (!inode_evicting)
8617 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8618 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8619 if (ordered) {
8620 /*
8621 * IO on this page will never be started, so we need
8622 * to account for any ordered extents now
8623 */
8624 if (!inode_evicting)
8625 clear_extent_bit(tree, page_start, page_end,
8626 EXTENT_DIRTY | EXTENT_DELALLOC |
8627 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8628 EXTENT_DEFRAG, 1, 0, &cached_state,
8629 GFP_NOFS);
8630 /*
8631 * whoever cleared the private bit is responsible
8632 * for the finish_ordered_io
8633 */
8634 if (TestClearPagePrivate2(page)) {
8635 struct btrfs_ordered_inode_tree *tree;
8636 u64 new_len;
8637
8638 tree = &BTRFS_I(inode)->ordered_tree;
8639
8640 spin_lock_irq(&tree->lock);
8641 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8642 new_len = page_start - ordered->file_offset;
8643 if (new_len < ordered->truncated_len)
8644 ordered->truncated_len = new_len;
8645 spin_unlock_irq(&tree->lock);
8646
8647 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8648 page_start,
8649 PAGE_CACHE_SIZE, 1))
8650 btrfs_finish_ordered_io(ordered);
8651 }
8652 btrfs_put_ordered_extent(ordered);
8653 if (!inode_evicting) {
8654 cached_state = NULL;
8655 lock_extent_bits(tree, page_start, page_end, 0,
8656 &cached_state);
8657 }
8658 }
8659
8660 /*
8661 * Qgroup reserved space handler
8662 * Page here will be either
8663 * 1) Already written to disk
8664 * In this case, its reserved space is released from data rsv map
8665 * and will be freed by delayed_ref handler finally.
8666 * So even we call qgroup_free_data(), it won't decrease reserved
8667 * space.
8668 * 2) Not written to disk
8669 * This means the reserved space should be freed here.
8670 */
8671 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8672 if (!inode_evicting) {
8673 clear_extent_bit(tree, page_start, page_end,
8674 EXTENT_LOCKED | EXTENT_DIRTY |
8675 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8676 EXTENT_DEFRAG, 1, 1,
8677 &cached_state, GFP_NOFS);
8678
8679 __btrfs_releasepage(page, GFP_NOFS);
8680 }
8681
8682 ClearPageChecked(page);
8683 if (PagePrivate(page)) {
8684 ClearPagePrivate(page);
8685 set_page_private(page, 0);
8686 page_cache_release(page);
8687 }
8688 }
8689
8690 /*
8691 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8692 * called from a page fault handler when a page is first dirtied. Hence we must
8693 * be careful to check for EOF conditions here. We set the page up correctly
8694 * for a written page which means we get ENOSPC checking when writing into
8695 * holes and correct delalloc and unwritten extent mapping on filesystems that
8696 * support these features.
8697 *
8698 * We are not allowed to take the i_mutex here so we have to play games to
8699 * protect against truncate races as the page could now be beyond EOF. Because
8700 * vmtruncate() writes the inode size before removing pages, once we have the
8701 * page lock we can determine safely if the page is beyond EOF. If it is not
8702 * beyond EOF, then the page is guaranteed safe against truncation until we
8703 * unlock the page.
8704 */
8705 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8706 {
8707 struct page *page = vmf->page;
8708 struct inode *inode = file_inode(vma->vm_file);
8709 struct btrfs_root *root = BTRFS_I(inode)->root;
8710 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8711 struct btrfs_ordered_extent *ordered;
8712 struct extent_state *cached_state = NULL;
8713 char *kaddr;
8714 unsigned long zero_start;
8715 loff_t size;
8716 int ret;
8717 int reserved = 0;
8718 u64 page_start;
8719 u64 page_end;
8720
8721 sb_start_pagefault(inode->i_sb);
8722 page_start = page_offset(page);
8723 page_end = page_start + PAGE_CACHE_SIZE - 1;
8724
8725 ret = btrfs_delalloc_reserve_space(inode, page_start,
8726 PAGE_CACHE_SIZE);
8727 if (!ret) {
8728 ret = file_update_time(vma->vm_file);
8729 reserved = 1;
8730 }
8731 if (ret) {
8732 if (ret == -ENOMEM)
8733 ret = VM_FAULT_OOM;
8734 else /* -ENOSPC, -EIO, etc */
8735 ret = VM_FAULT_SIGBUS;
8736 if (reserved)
8737 goto out;
8738 goto out_noreserve;
8739 }
8740
8741 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8742 again:
8743 lock_page(page);
8744 size = i_size_read(inode);
8745
8746 if ((page->mapping != inode->i_mapping) ||
8747 (page_start >= size)) {
8748 /* page got truncated out from underneath us */
8749 goto out_unlock;
8750 }
8751 wait_on_page_writeback(page);
8752
8753 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8754 set_page_extent_mapped(page);
8755
8756 /*
8757 * we can't set the delalloc bits if there are pending ordered
8758 * extents. Drop our locks and wait for them to finish
8759 */
8760 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8761 if (ordered) {
8762 unlock_extent_cached(io_tree, page_start, page_end,
8763 &cached_state, GFP_NOFS);
8764 unlock_page(page);
8765 btrfs_start_ordered_extent(inode, ordered, 1);
8766 btrfs_put_ordered_extent(ordered);
8767 goto again;
8768 }
8769
8770 /*
8771 * XXX - page_mkwrite gets called every time the page is dirtied, even
8772 * if it was already dirty, so for space accounting reasons we need to
8773 * clear any delalloc bits for the range we are fixing to save. There
8774 * is probably a better way to do this, but for now keep consistent with
8775 * prepare_pages in the normal write path.
8776 */
8777 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8778 EXTENT_DIRTY | EXTENT_DELALLOC |
8779 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8780 0, 0, &cached_state, GFP_NOFS);
8781
8782 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8783 &cached_state);
8784 if (ret) {
8785 unlock_extent_cached(io_tree, page_start, page_end,
8786 &cached_state, GFP_NOFS);
8787 ret = VM_FAULT_SIGBUS;
8788 goto out_unlock;
8789 }
8790 ret = 0;
8791
8792 /* page is wholly or partially inside EOF */
8793 if (page_start + PAGE_CACHE_SIZE > size)
8794 zero_start = size & ~PAGE_CACHE_MASK;
8795 else
8796 zero_start = PAGE_CACHE_SIZE;
8797
8798 if (zero_start != PAGE_CACHE_SIZE) {
8799 kaddr = kmap(page);
8800 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8801 flush_dcache_page(page);
8802 kunmap(page);
8803 }
8804 ClearPageChecked(page);
8805 set_page_dirty(page);
8806 SetPageUptodate(page);
8807
8808 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8809 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8810 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8811
8812 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8813
8814 out_unlock:
8815 if (!ret) {
8816 sb_end_pagefault(inode->i_sb);
8817 return VM_FAULT_LOCKED;
8818 }
8819 unlock_page(page);
8820 out:
8821 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8822 out_noreserve:
8823 sb_end_pagefault(inode->i_sb);
8824 return ret;
8825 }
8826
8827 static int btrfs_truncate(struct inode *inode)
8828 {
8829 struct btrfs_root *root = BTRFS_I(inode)->root;
8830 struct btrfs_block_rsv *rsv;
8831 int ret = 0;
8832 int err = 0;
8833 struct btrfs_trans_handle *trans;
8834 u64 mask = root->sectorsize - 1;
8835 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8836
8837 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8838 (u64)-1);
8839 if (ret)
8840 return ret;
8841
8842 /*
8843 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8844 * 3 things going on here
8845 *
8846 * 1) We need to reserve space for our orphan item and the space to
8847 * delete our orphan item. Lord knows we don't want to have a dangling
8848 * orphan item because we didn't reserve space to remove it.
8849 *
8850 * 2) We need to reserve space to update our inode.
8851 *
8852 * 3) We need to have something to cache all the space that is going to
8853 * be free'd up by the truncate operation, but also have some slack
8854 * space reserved in case it uses space during the truncate (thank you
8855 * very much snapshotting).
8856 *
8857 * And we need these to all be seperate. The fact is we can use alot of
8858 * space doing the truncate, and we have no earthly idea how much space
8859 * we will use, so we need the truncate reservation to be seperate so it
8860 * doesn't end up using space reserved for updating the inode or
8861 * removing the orphan item. We also need to be able to stop the
8862 * transaction and start a new one, which means we need to be able to
8863 * update the inode several times, and we have no idea of knowing how
8864 * many times that will be, so we can't just reserve 1 item for the
8865 * entirety of the opration, so that has to be done seperately as well.
8866 * Then there is the orphan item, which does indeed need to be held on
8867 * to for the whole operation, and we need nobody to touch this reserved
8868 * space except the orphan code.
8869 *
8870 * So that leaves us with
8871 *
8872 * 1) root->orphan_block_rsv - for the orphan deletion.
8873 * 2) rsv - for the truncate reservation, which we will steal from the
8874 * transaction reservation.
8875 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8876 * updating the inode.
8877 */
8878 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8879 if (!rsv)
8880 return -ENOMEM;
8881 rsv->size = min_size;
8882 rsv->failfast = 1;
8883
8884 /*
8885 * 1 for the truncate slack space
8886 * 1 for updating the inode.
8887 */
8888 trans = btrfs_start_transaction(root, 2);
8889 if (IS_ERR(trans)) {
8890 err = PTR_ERR(trans);
8891 goto out;
8892 }
8893
8894 /* Migrate the slack space for the truncate to our reserve */
8895 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8896 min_size);
8897 BUG_ON(ret);
8898
8899 /*
8900 * So if we truncate and then write and fsync we normally would just
8901 * write the extents that changed, which is a problem if we need to
8902 * first truncate that entire inode. So set this flag so we write out
8903 * all of the extents in the inode to the sync log so we're completely
8904 * safe.
8905 */
8906 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8907 trans->block_rsv = rsv;
8908
8909 while (1) {
8910 ret = btrfs_truncate_inode_items(trans, root, inode,
8911 inode->i_size,
8912 BTRFS_EXTENT_DATA_KEY);
8913 if (ret != -ENOSPC && ret != -EAGAIN) {
8914 err = ret;
8915 break;
8916 }
8917
8918 trans->block_rsv = &root->fs_info->trans_block_rsv;
8919 ret = btrfs_update_inode(trans, root, inode);
8920 if (ret) {
8921 err = ret;
8922 break;
8923 }
8924
8925 btrfs_end_transaction(trans, root);
8926 btrfs_btree_balance_dirty(root);
8927
8928 trans = btrfs_start_transaction(root, 2);
8929 if (IS_ERR(trans)) {
8930 ret = err = PTR_ERR(trans);
8931 trans = NULL;
8932 break;
8933 }
8934
8935 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8936 rsv, min_size);
8937 BUG_ON(ret); /* shouldn't happen */
8938 trans->block_rsv = rsv;
8939 }
8940
8941 if (ret == 0 && inode->i_nlink > 0) {
8942 trans->block_rsv = root->orphan_block_rsv;
8943 ret = btrfs_orphan_del(trans, inode);
8944 if (ret)
8945 err = ret;
8946 }
8947
8948 if (trans) {
8949 trans->block_rsv = &root->fs_info->trans_block_rsv;
8950 ret = btrfs_update_inode(trans, root, inode);
8951 if (ret && !err)
8952 err = ret;
8953
8954 ret = btrfs_end_transaction(trans, root);
8955 btrfs_btree_balance_dirty(root);
8956 }
8957
8958 out:
8959 btrfs_free_block_rsv(root, rsv);
8960
8961 if (ret && !err)
8962 err = ret;
8963
8964 return err;
8965 }
8966
8967 /*
8968 * create a new subvolume directory/inode (helper for the ioctl).
8969 */
8970 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8971 struct btrfs_root *new_root,
8972 struct btrfs_root *parent_root,
8973 u64 new_dirid)
8974 {
8975 struct inode *inode;
8976 int err;
8977 u64 index = 0;
8978
8979 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8980 new_dirid, new_dirid,
8981 S_IFDIR | (~current_umask() & S_IRWXUGO),
8982 &index);
8983 if (IS_ERR(inode))
8984 return PTR_ERR(inode);
8985 inode->i_op = &btrfs_dir_inode_operations;
8986 inode->i_fop = &btrfs_dir_file_operations;
8987
8988 set_nlink(inode, 1);
8989 btrfs_i_size_write(inode, 0);
8990 unlock_new_inode(inode);
8991
8992 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8993 if (err)
8994 btrfs_err(new_root->fs_info,
8995 "error inheriting subvolume %llu properties: %d",
8996 new_root->root_key.objectid, err);
8997
8998 err = btrfs_update_inode(trans, new_root, inode);
8999
9000 iput(inode);
9001 return err;
9002 }
9003
9004 struct inode *btrfs_alloc_inode(struct super_block *sb)
9005 {
9006 struct btrfs_inode *ei;
9007 struct inode *inode;
9008
9009 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9010 if (!ei)
9011 return NULL;
9012
9013 ei->root = NULL;
9014 ei->generation = 0;
9015 ei->last_trans = 0;
9016 ei->last_sub_trans = 0;
9017 ei->logged_trans = 0;
9018 ei->delalloc_bytes = 0;
9019 ei->defrag_bytes = 0;
9020 ei->disk_i_size = 0;
9021 ei->flags = 0;
9022 ei->csum_bytes = 0;
9023 ei->index_cnt = (u64)-1;
9024 ei->dir_index = 0;
9025 ei->last_unlink_trans = 0;
9026 ei->last_log_commit = 0;
9027
9028 spin_lock_init(&ei->lock);
9029 ei->outstanding_extents = 0;
9030 ei->reserved_extents = 0;
9031
9032 ei->runtime_flags = 0;
9033 ei->force_compress = BTRFS_COMPRESS_NONE;
9034
9035 ei->delayed_node = NULL;
9036
9037 ei->i_otime.tv_sec = 0;
9038 ei->i_otime.tv_nsec = 0;
9039
9040 inode = &ei->vfs_inode;
9041 extent_map_tree_init(&ei->extent_tree);
9042 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9043 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9044 ei->io_tree.track_uptodate = 1;
9045 ei->io_failure_tree.track_uptodate = 1;
9046 atomic_set(&ei->sync_writers, 0);
9047 mutex_init(&ei->log_mutex);
9048 mutex_init(&ei->delalloc_mutex);
9049 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9050 INIT_LIST_HEAD(&ei->delalloc_inodes);
9051 RB_CLEAR_NODE(&ei->rb_node);
9052
9053 return inode;
9054 }
9055
9056 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9057 void btrfs_test_destroy_inode(struct inode *inode)
9058 {
9059 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9060 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9061 }
9062 #endif
9063
9064 static void btrfs_i_callback(struct rcu_head *head)
9065 {
9066 struct inode *inode = container_of(head, struct inode, i_rcu);
9067 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9068 }
9069
9070 void btrfs_destroy_inode(struct inode *inode)
9071 {
9072 struct btrfs_ordered_extent *ordered;
9073 struct btrfs_root *root = BTRFS_I(inode)->root;
9074
9075 WARN_ON(!hlist_empty(&inode->i_dentry));
9076 WARN_ON(inode->i_data.nrpages);
9077 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9078 WARN_ON(BTRFS_I(inode)->reserved_extents);
9079 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9080 WARN_ON(BTRFS_I(inode)->csum_bytes);
9081 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9082
9083 /*
9084 * This can happen where we create an inode, but somebody else also
9085 * created the same inode and we need to destroy the one we already
9086 * created.
9087 */
9088 if (!root)
9089 goto free;
9090
9091 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9092 &BTRFS_I(inode)->runtime_flags)) {
9093 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9094 btrfs_ino(inode));
9095 atomic_dec(&root->orphan_inodes);
9096 }
9097
9098 while (1) {
9099 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9100 if (!ordered)
9101 break;
9102 else {
9103 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9104 ordered->file_offset, ordered->len);
9105 btrfs_remove_ordered_extent(inode, ordered);
9106 btrfs_put_ordered_extent(ordered);
9107 btrfs_put_ordered_extent(ordered);
9108 }
9109 }
9110 btrfs_qgroup_check_reserved_leak(inode);
9111 inode_tree_del(inode);
9112 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9113 free:
9114 call_rcu(&inode->i_rcu, btrfs_i_callback);
9115 }
9116
9117 int btrfs_drop_inode(struct inode *inode)
9118 {
9119 struct btrfs_root *root = BTRFS_I(inode)->root;
9120
9121 if (root == NULL)
9122 return 1;
9123
9124 /* the snap/subvol tree is on deleting */
9125 if (btrfs_root_refs(&root->root_item) == 0)
9126 return 1;
9127 else
9128 return generic_drop_inode(inode);
9129 }
9130
9131 static void init_once(void *foo)
9132 {
9133 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9134
9135 inode_init_once(&ei->vfs_inode);
9136 }
9137
9138 void btrfs_destroy_cachep(void)
9139 {
9140 /*
9141 * Make sure all delayed rcu free inodes are flushed before we
9142 * destroy cache.
9143 */
9144 rcu_barrier();
9145 if (btrfs_inode_cachep)
9146 kmem_cache_destroy(btrfs_inode_cachep);
9147 if (btrfs_trans_handle_cachep)
9148 kmem_cache_destroy(btrfs_trans_handle_cachep);
9149 if (btrfs_transaction_cachep)
9150 kmem_cache_destroy(btrfs_transaction_cachep);
9151 if (btrfs_path_cachep)
9152 kmem_cache_destroy(btrfs_path_cachep);
9153 if (btrfs_free_space_cachep)
9154 kmem_cache_destroy(btrfs_free_space_cachep);
9155 if (btrfs_delalloc_work_cachep)
9156 kmem_cache_destroy(btrfs_delalloc_work_cachep);
9157 }
9158
9159 int btrfs_init_cachep(void)
9160 {
9161 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9162 sizeof(struct btrfs_inode), 0,
9163 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9164 if (!btrfs_inode_cachep)
9165 goto fail;
9166
9167 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9168 sizeof(struct btrfs_trans_handle), 0,
9169 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9170 if (!btrfs_trans_handle_cachep)
9171 goto fail;
9172
9173 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9174 sizeof(struct btrfs_transaction), 0,
9175 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9176 if (!btrfs_transaction_cachep)
9177 goto fail;
9178
9179 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9180 sizeof(struct btrfs_path), 0,
9181 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9182 if (!btrfs_path_cachep)
9183 goto fail;
9184
9185 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9186 sizeof(struct btrfs_free_space), 0,
9187 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9188 if (!btrfs_free_space_cachep)
9189 goto fail;
9190
9191 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9192 sizeof(struct btrfs_delalloc_work), 0,
9193 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9194 NULL);
9195 if (!btrfs_delalloc_work_cachep)
9196 goto fail;
9197
9198 return 0;
9199 fail:
9200 btrfs_destroy_cachep();
9201 return -ENOMEM;
9202 }
9203
9204 static int btrfs_getattr(struct vfsmount *mnt,
9205 struct dentry *dentry, struct kstat *stat)
9206 {
9207 u64 delalloc_bytes;
9208 struct inode *inode = d_inode(dentry);
9209 u32 blocksize = inode->i_sb->s_blocksize;
9210
9211 generic_fillattr(inode, stat);
9212 stat->dev = BTRFS_I(inode)->root->anon_dev;
9213 stat->blksize = PAGE_CACHE_SIZE;
9214
9215 spin_lock(&BTRFS_I(inode)->lock);
9216 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9217 spin_unlock(&BTRFS_I(inode)->lock);
9218 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9219 ALIGN(delalloc_bytes, blocksize)) >> 9;
9220 return 0;
9221 }
9222
9223 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9224 struct inode *new_dir, struct dentry *new_dentry)
9225 {
9226 struct btrfs_trans_handle *trans;
9227 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9228 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9229 struct inode *new_inode = d_inode(new_dentry);
9230 struct inode *old_inode = d_inode(old_dentry);
9231 struct timespec ctime = CURRENT_TIME;
9232 u64 index = 0;
9233 u64 root_objectid;
9234 int ret;
9235 u64 old_ino = btrfs_ino(old_inode);
9236
9237 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9238 return -EPERM;
9239
9240 /* we only allow rename subvolume link between subvolumes */
9241 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9242 return -EXDEV;
9243
9244 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9245 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9246 return -ENOTEMPTY;
9247
9248 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9249 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9250 return -ENOTEMPTY;
9251
9252
9253 /* check for collisions, even if the name isn't there */
9254 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9255 new_dentry->d_name.name,
9256 new_dentry->d_name.len);
9257
9258 if (ret) {
9259 if (ret == -EEXIST) {
9260 /* we shouldn't get
9261 * eexist without a new_inode */
9262 if (WARN_ON(!new_inode)) {
9263 return ret;
9264 }
9265 } else {
9266 /* maybe -EOVERFLOW */
9267 return ret;
9268 }
9269 }
9270 ret = 0;
9271
9272 /*
9273 * we're using rename to replace one file with another. Start IO on it
9274 * now so we don't add too much work to the end of the transaction
9275 */
9276 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9277 filemap_flush(old_inode->i_mapping);
9278
9279 /* close the racy window with snapshot create/destroy ioctl */
9280 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9281 down_read(&root->fs_info->subvol_sem);
9282 /*
9283 * We want to reserve the absolute worst case amount of items. So if
9284 * both inodes are subvols and we need to unlink them then that would
9285 * require 4 item modifications, but if they are both normal inodes it
9286 * would require 5 item modifications, so we'll assume their normal
9287 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9288 * should cover the worst case number of items we'll modify.
9289 */
9290 trans = btrfs_start_transaction(root, 11);
9291 if (IS_ERR(trans)) {
9292 ret = PTR_ERR(trans);
9293 goto out_notrans;
9294 }
9295
9296 if (dest != root)
9297 btrfs_record_root_in_trans(trans, dest);
9298
9299 ret = btrfs_set_inode_index(new_dir, &index);
9300 if (ret)
9301 goto out_fail;
9302
9303 BTRFS_I(old_inode)->dir_index = 0ULL;
9304 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9305 /* force full log commit if subvolume involved. */
9306 btrfs_set_log_full_commit(root->fs_info, trans);
9307 } else {
9308 ret = btrfs_insert_inode_ref(trans, dest,
9309 new_dentry->d_name.name,
9310 new_dentry->d_name.len,
9311 old_ino,
9312 btrfs_ino(new_dir), index);
9313 if (ret)
9314 goto out_fail;
9315 /*
9316 * this is an ugly little race, but the rename is required
9317 * to make sure that if we crash, the inode is either at the
9318 * old name or the new one. pinning the log transaction lets
9319 * us make sure we don't allow a log commit to come in after
9320 * we unlink the name but before we add the new name back in.
9321 */
9322 btrfs_pin_log_trans(root);
9323 }
9324
9325 inode_inc_iversion(old_dir);
9326 inode_inc_iversion(new_dir);
9327 inode_inc_iversion(old_inode);
9328 old_dir->i_ctime = old_dir->i_mtime = ctime;
9329 new_dir->i_ctime = new_dir->i_mtime = ctime;
9330 old_inode->i_ctime = ctime;
9331
9332 if (old_dentry->d_parent != new_dentry->d_parent)
9333 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9334
9335 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9336 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9337 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9338 old_dentry->d_name.name,
9339 old_dentry->d_name.len);
9340 } else {
9341 ret = __btrfs_unlink_inode(trans, root, old_dir,
9342 d_inode(old_dentry),
9343 old_dentry->d_name.name,
9344 old_dentry->d_name.len);
9345 if (!ret)
9346 ret = btrfs_update_inode(trans, root, old_inode);
9347 }
9348 if (ret) {
9349 btrfs_abort_transaction(trans, root, ret);
9350 goto out_fail;
9351 }
9352
9353 if (new_inode) {
9354 inode_inc_iversion(new_inode);
9355 new_inode->i_ctime = CURRENT_TIME;
9356 if (unlikely(btrfs_ino(new_inode) ==
9357 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9358 root_objectid = BTRFS_I(new_inode)->location.objectid;
9359 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9360 root_objectid,
9361 new_dentry->d_name.name,
9362 new_dentry->d_name.len);
9363 BUG_ON(new_inode->i_nlink == 0);
9364 } else {
9365 ret = btrfs_unlink_inode(trans, dest, new_dir,
9366 d_inode(new_dentry),
9367 new_dentry->d_name.name,
9368 new_dentry->d_name.len);
9369 }
9370 if (!ret && new_inode->i_nlink == 0)
9371 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9372 if (ret) {
9373 btrfs_abort_transaction(trans, root, ret);
9374 goto out_fail;
9375 }
9376 }
9377
9378 ret = btrfs_add_link(trans, new_dir, old_inode,
9379 new_dentry->d_name.name,
9380 new_dentry->d_name.len, 0, index);
9381 if (ret) {
9382 btrfs_abort_transaction(trans, root, ret);
9383 goto out_fail;
9384 }
9385
9386 if (old_inode->i_nlink == 1)
9387 BTRFS_I(old_inode)->dir_index = index;
9388
9389 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9390 struct dentry *parent = new_dentry->d_parent;
9391 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9392 btrfs_end_log_trans(root);
9393 }
9394 out_fail:
9395 btrfs_end_transaction(trans, root);
9396 out_notrans:
9397 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9398 up_read(&root->fs_info->subvol_sem);
9399
9400 return ret;
9401 }
9402
9403 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9404 struct inode *new_dir, struct dentry *new_dentry,
9405 unsigned int flags)
9406 {
9407 if (flags & ~RENAME_NOREPLACE)
9408 return -EINVAL;
9409
9410 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9411 }
9412
9413 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9414 {
9415 struct btrfs_delalloc_work *delalloc_work;
9416 struct inode *inode;
9417
9418 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9419 work);
9420 inode = delalloc_work->inode;
9421 if (delalloc_work->wait) {
9422 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9423 } else {
9424 filemap_flush(inode->i_mapping);
9425 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9426 &BTRFS_I(inode)->runtime_flags))
9427 filemap_flush(inode->i_mapping);
9428 }
9429
9430 if (delalloc_work->delay_iput)
9431 btrfs_add_delayed_iput(inode);
9432 else
9433 iput(inode);
9434 complete(&delalloc_work->completion);
9435 }
9436
9437 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9438 int wait, int delay_iput)
9439 {
9440 struct btrfs_delalloc_work *work;
9441
9442 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9443 if (!work)
9444 return NULL;
9445
9446 init_completion(&work->completion);
9447 INIT_LIST_HEAD(&work->list);
9448 work->inode = inode;
9449 work->wait = wait;
9450 work->delay_iput = delay_iput;
9451 WARN_ON_ONCE(!inode);
9452 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9453 btrfs_run_delalloc_work, NULL, NULL);
9454
9455 return work;
9456 }
9457
9458 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9459 {
9460 wait_for_completion(&work->completion);
9461 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9462 }
9463
9464 /*
9465 * some fairly slow code that needs optimization. This walks the list
9466 * of all the inodes with pending delalloc and forces them to disk.
9467 */
9468 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9469 int nr)
9470 {
9471 struct btrfs_inode *binode;
9472 struct inode *inode;
9473 struct btrfs_delalloc_work *work, *next;
9474 struct list_head works;
9475 struct list_head splice;
9476 int ret = 0;
9477
9478 INIT_LIST_HEAD(&works);
9479 INIT_LIST_HEAD(&splice);
9480
9481 mutex_lock(&root->delalloc_mutex);
9482 spin_lock(&root->delalloc_lock);
9483 list_splice_init(&root->delalloc_inodes, &splice);
9484 while (!list_empty(&splice)) {
9485 binode = list_entry(splice.next, struct btrfs_inode,
9486 delalloc_inodes);
9487
9488 list_move_tail(&binode->delalloc_inodes,
9489 &root->delalloc_inodes);
9490 inode = igrab(&binode->vfs_inode);
9491 if (!inode) {
9492 cond_resched_lock(&root->delalloc_lock);
9493 continue;
9494 }
9495 spin_unlock(&root->delalloc_lock);
9496
9497 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9498 if (!work) {
9499 if (delay_iput)
9500 btrfs_add_delayed_iput(inode);
9501 else
9502 iput(inode);
9503 ret = -ENOMEM;
9504 goto out;
9505 }
9506 list_add_tail(&work->list, &works);
9507 btrfs_queue_work(root->fs_info->flush_workers,
9508 &work->work);
9509 ret++;
9510 if (nr != -1 && ret >= nr)
9511 goto out;
9512 cond_resched();
9513 spin_lock(&root->delalloc_lock);
9514 }
9515 spin_unlock(&root->delalloc_lock);
9516
9517 out:
9518 list_for_each_entry_safe(work, next, &works, list) {
9519 list_del_init(&work->list);
9520 btrfs_wait_and_free_delalloc_work(work);
9521 }
9522
9523 if (!list_empty_careful(&splice)) {
9524 spin_lock(&root->delalloc_lock);
9525 list_splice_tail(&splice, &root->delalloc_inodes);
9526 spin_unlock(&root->delalloc_lock);
9527 }
9528 mutex_unlock(&root->delalloc_mutex);
9529 return ret;
9530 }
9531
9532 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9533 {
9534 int ret;
9535
9536 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9537 return -EROFS;
9538
9539 ret = __start_delalloc_inodes(root, delay_iput, -1);
9540 if (ret > 0)
9541 ret = 0;
9542 /*
9543 * the filemap_flush will queue IO into the worker threads, but
9544 * we have to make sure the IO is actually started and that
9545 * ordered extents get created before we return
9546 */
9547 atomic_inc(&root->fs_info->async_submit_draining);
9548 while (atomic_read(&root->fs_info->nr_async_submits) ||
9549 atomic_read(&root->fs_info->async_delalloc_pages)) {
9550 wait_event(root->fs_info->async_submit_wait,
9551 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9552 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9553 }
9554 atomic_dec(&root->fs_info->async_submit_draining);
9555 return ret;
9556 }
9557
9558 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9559 int nr)
9560 {
9561 struct btrfs_root *root;
9562 struct list_head splice;
9563 int ret;
9564
9565 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9566 return -EROFS;
9567
9568 INIT_LIST_HEAD(&splice);
9569
9570 mutex_lock(&fs_info->delalloc_root_mutex);
9571 spin_lock(&fs_info->delalloc_root_lock);
9572 list_splice_init(&fs_info->delalloc_roots, &splice);
9573 while (!list_empty(&splice) && nr) {
9574 root = list_first_entry(&splice, struct btrfs_root,
9575 delalloc_root);
9576 root = btrfs_grab_fs_root(root);
9577 BUG_ON(!root);
9578 list_move_tail(&root->delalloc_root,
9579 &fs_info->delalloc_roots);
9580 spin_unlock(&fs_info->delalloc_root_lock);
9581
9582 ret = __start_delalloc_inodes(root, delay_iput, nr);
9583 btrfs_put_fs_root(root);
9584 if (ret < 0)
9585 goto out;
9586
9587 if (nr != -1) {
9588 nr -= ret;
9589 WARN_ON(nr < 0);
9590 }
9591 spin_lock(&fs_info->delalloc_root_lock);
9592 }
9593 spin_unlock(&fs_info->delalloc_root_lock);
9594
9595 ret = 0;
9596 atomic_inc(&fs_info->async_submit_draining);
9597 while (atomic_read(&fs_info->nr_async_submits) ||
9598 atomic_read(&fs_info->async_delalloc_pages)) {
9599 wait_event(fs_info->async_submit_wait,
9600 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9601 atomic_read(&fs_info->async_delalloc_pages) == 0));
9602 }
9603 atomic_dec(&fs_info->async_submit_draining);
9604 out:
9605 if (!list_empty_careful(&splice)) {
9606 spin_lock(&fs_info->delalloc_root_lock);
9607 list_splice_tail(&splice, &fs_info->delalloc_roots);
9608 spin_unlock(&fs_info->delalloc_root_lock);
9609 }
9610 mutex_unlock(&fs_info->delalloc_root_mutex);
9611 return ret;
9612 }
9613
9614 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9615 const char *symname)
9616 {
9617 struct btrfs_trans_handle *trans;
9618 struct btrfs_root *root = BTRFS_I(dir)->root;
9619 struct btrfs_path *path;
9620 struct btrfs_key key;
9621 struct inode *inode = NULL;
9622 int err;
9623 int drop_inode = 0;
9624 u64 objectid;
9625 u64 index = 0;
9626 int name_len;
9627 int datasize;
9628 unsigned long ptr;
9629 struct btrfs_file_extent_item *ei;
9630 struct extent_buffer *leaf;
9631
9632 name_len = strlen(symname);
9633 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9634 return -ENAMETOOLONG;
9635
9636 /*
9637 * 2 items for inode item and ref
9638 * 2 items for dir items
9639 * 1 item for xattr if selinux is on
9640 */
9641 trans = btrfs_start_transaction(root, 5);
9642 if (IS_ERR(trans))
9643 return PTR_ERR(trans);
9644
9645 err = btrfs_find_free_ino(root, &objectid);
9646 if (err)
9647 goto out_unlock;
9648
9649 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9650 dentry->d_name.len, btrfs_ino(dir), objectid,
9651 S_IFLNK|S_IRWXUGO, &index);
9652 if (IS_ERR(inode)) {
9653 err = PTR_ERR(inode);
9654 goto out_unlock;
9655 }
9656
9657 /*
9658 * If the active LSM wants to access the inode during
9659 * d_instantiate it needs these. Smack checks to see
9660 * if the filesystem supports xattrs by looking at the
9661 * ops vector.
9662 */
9663 inode->i_fop = &btrfs_file_operations;
9664 inode->i_op = &btrfs_file_inode_operations;
9665 inode->i_mapping->a_ops = &btrfs_aops;
9666 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9667
9668 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9669 if (err)
9670 goto out_unlock_inode;
9671
9672 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9673 if (err)
9674 goto out_unlock_inode;
9675
9676 path = btrfs_alloc_path();
9677 if (!path) {
9678 err = -ENOMEM;
9679 goto out_unlock_inode;
9680 }
9681 key.objectid = btrfs_ino(inode);
9682 key.offset = 0;
9683 key.type = BTRFS_EXTENT_DATA_KEY;
9684 datasize = btrfs_file_extent_calc_inline_size(name_len);
9685 err = btrfs_insert_empty_item(trans, root, path, &key,
9686 datasize);
9687 if (err) {
9688 btrfs_free_path(path);
9689 goto out_unlock_inode;
9690 }
9691 leaf = path->nodes[0];
9692 ei = btrfs_item_ptr(leaf, path->slots[0],
9693 struct btrfs_file_extent_item);
9694 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9695 btrfs_set_file_extent_type(leaf, ei,
9696 BTRFS_FILE_EXTENT_INLINE);
9697 btrfs_set_file_extent_encryption(leaf, ei, 0);
9698 btrfs_set_file_extent_compression(leaf, ei, 0);
9699 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9700 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9701
9702 ptr = btrfs_file_extent_inline_start(ei);
9703 write_extent_buffer(leaf, symname, ptr, name_len);
9704 btrfs_mark_buffer_dirty(leaf);
9705 btrfs_free_path(path);
9706
9707 inode->i_op = &btrfs_symlink_inode_operations;
9708 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9709 inode_set_bytes(inode, name_len);
9710 btrfs_i_size_write(inode, name_len);
9711 err = btrfs_update_inode(trans, root, inode);
9712 if (err) {
9713 drop_inode = 1;
9714 goto out_unlock_inode;
9715 }
9716
9717 unlock_new_inode(inode);
9718 d_instantiate(dentry, inode);
9719
9720 out_unlock:
9721 btrfs_end_transaction(trans, root);
9722 if (drop_inode) {
9723 inode_dec_link_count(inode);
9724 iput(inode);
9725 }
9726 btrfs_btree_balance_dirty(root);
9727 return err;
9728
9729 out_unlock_inode:
9730 drop_inode = 1;
9731 unlock_new_inode(inode);
9732 goto out_unlock;
9733 }
9734
9735 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9736 u64 start, u64 num_bytes, u64 min_size,
9737 loff_t actual_len, u64 *alloc_hint,
9738 struct btrfs_trans_handle *trans)
9739 {
9740 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9741 struct extent_map *em;
9742 struct btrfs_root *root = BTRFS_I(inode)->root;
9743 struct btrfs_key ins;
9744 u64 cur_offset = start;
9745 u64 i_size;
9746 u64 cur_bytes;
9747 u64 last_alloc = (u64)-1;
9748 int ret = 0;
9749 bool own_trans = true;
9750
9751 if (trans)
9752 own_trans = false;
9753 while (num_bytes > 0) {
9754 if (own_trans) {
9755 trans = btrfs_start_transaction(root, 3);
9756 if (IS_ERR(trans)) {
9757 ret = PTR_ERR(trans);
9758 break;
9759 }
9760 }
9761
9762 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9763 cur_bytes = max(cur_bytes, min_size);
9764 /*
9765 * If we are severely fragmented we could end up with really
9766 * small allocations, so if the allocator is returning small
9767 * chunks lets make its job easier by only searching for those
9768 * sized chunks.
9769 */
9770 cur_bytes = min(cur_bytes, last_alloc);
9771 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9772 *alloc_hint, &ins, 1, 0);
9773 if (ret) {
9774 if (own_trans)
9775 btrfs_end_transaction(trans, root);
9776 break;
9777 }
9778
9779 last_alloc = ins.offset;
9780 ret = insert_reserved_file_extent(trans, inode,
9781 cur_offset, ins.objectid,
9782 ins.offset, ins.offset,
9783 ins.offset, 0, 0, 0,
9784 BTRFS_FILE_EXTENT_PREALLOC);
9785 if (ret) {
9786 btrfs_free_reserved_extent(root, ins.objectid,
9787 ins.offset, 0);
9788 btrfs_abort_transaction(trans, root, ret);
9789 if (own_trans)
9790 btrfs_end_transaction(trans, root);
9791 break;
9792 }
9793
9794 btrfs_drop_extent_cache(inode, cur_offset,
9795 cur_offset + ins.offset -1, 0);
9796
9797 em = alloc_extent_map();
9798 if (!em) {
9799 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9800 &BTRFS_I(inode)->runtime_flags);
9801 goto next;
9802 }
9803
9804 em->start = cur_offset;
9805 em->orig_start = cur_offset;
9806 em->len = ins.offset;
9807 em->block_start = ins.objectid;
9808 em->block_len = ins.offset;
9809 em->orig_block_len = ins.offset;
9810 em->ram_bytes = ins.offset;
9811 em->bdev = root->fs_info->fs_devices->latest_bdev;
9812 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9813 em->generation = trans->transid;
9814
9815 while (1) {
9816 write_lock(&em_tree->lock);
9817 ret = add_extent_mapping(em_tree, em, 1);
9818 write_unlock(&em_tree->lock);
9819 if (ret != -EEXIST)
9820 break;
9821 btrfs_drop_extent_cache(inode, cur_offset,
9822 cur_offset + ins.offset - 1,
9823 0);
9824 }
9825 free_extent_map(em);
9826 next:
9827 num_bytes -= ins.offset;
9828 cur_offset += ins.offset;
9829 *alloc_hint = ins.objectid + ins.offset;
9830
9831 inode_inc_iversion(inode);
9832 inode->i_ctime = CURRENT_TIME;
9833 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9834 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9835 (actual_len > inode->i_size) &&
9836 (cur_offset > inode->i_size)) {
9837 if (cur_offset > actual_len)
9838 i_size = actual_len;
9839 else
9840 i_size = cur_offset;
9841 i_size_write(inode, i_size);
9842 btrfs_ordered_update_i_size(inode, i_size, NULL);
9843 }
9844
9845 ret = btrfs_update_inode(trans, root, inode);
9846
9847 if (ret) {
9848 btrfs_abort_transaction(trans, root, ret);
9849 if (own_trans)
9850 btrfs_end_transaction(trans, root);
9851 break;
9852 }
9853
9854 if (own_trans)
9855 btrfs_end_transaction(trans, root);
9856 }
9857 return ret;
9858 }
9859
9860 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9861 u64 start, u64 num_bytes, u64 min_size,
9862 loff_t actual_len, u64 *alloc_hint)
9863 {
9864 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9865 min_size, actual_len, alloc_hint,
9866 NULL);
9867 }
9868
9869 int btrfs_prealloc_file_range_trans(struct inode *inode,
9870 struct btrfs_trans_handle *trans, int mode,
9871 u64 start, u64 num_bytes, u64 min_size,
9872 loff_t actual_len, u64 *alloc_hint)
9873 {
9874 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9875 min_size, actual_len, alloc_hint, trans);
9876 }
9877
9878 static int btrfs_set_page_dirty(struct page *page)
9879 {
9880 return __set_page_dirty_nobuffers(page);
9881 }
9882
9883 static int btrfs_permission(struct inode *inode, int mask)
9884 {
9885 struct btrfs_root *root = BTRFS_I(inode)->root;
9886 umode_t mode = inode->i_mode;
9887
9888 if (mask & MAY_WRITE &&
9889 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9890 if (btrfs_root_readonly(root))
9891 return -EROFS;
9892 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9893 return -EACCES;
9894 }
9895 return generic_permission(inode, mask);
9896 }
9897
9898 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9899 {
9900 struct btrfs_trans_handle *trans;
9901 struct btrfs_root *root = BTRFS_I(dir)->root;
9902 struct inode *inode = NULL;
9903 u64 objectid;
9904 u64 index;
9905 int ret = 0;
9906
9907 /*
9908 * 5 units required for adding orphan entry
9909 */
9910 trans = btrfs_start_transaction(root, 5);
9911 if (IS_ERR(trans))
9912 return PTR_ERR(trans);
9913
9914 ret = btrfs_find_free_ino(root, &objectid);
9915 if (ret)
9916 goto out;
9917
9918 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9919 btrfs_ino(dir), objectid, mode, &index);
9920 if (IS_ERR(inode)) {
9921 ret = PTR_ERR(inode);
9922 inode = NULL;
9923 goto out;
9924 }
9925
9926 inode->i_fop = &btrfs_file_operations;
9927 inode->i_op = &btrfs_file_inode_operations;
9928
9929 inode->i_mapping->a_ops = &btrfs_aops;
9930 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9931
9932 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9933 if (ret)
9934 goto out_inode;
9935
9936 ret = btrfs_update_inode(trans, root, inode);
9937 if (ret)
9938 goto out_inode;
9939 ret = btrfs_orphan_add(trans, inode);
9940 if (ret)
9941 goto out_inode;
9942
9943 /*
9944 * We set number of links to 0 in btrfs_new_inode(), and here we set
9945 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9946 * through:
9947 *
9948 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9949 */
9950 set_nlink(inode, 1);
9951 unlock_new_inode(inode);
9952 d_tmpfile(dentry, inode);
9953 mark_inode_dirty(inode);
9954
9955 out:
9956 btrfs_end_transaction(trans, root);
9957 if (ret)
9958 iput(inode);
9959 btrfs_balance_delayed_items(root);
9960 btrfs_btree_balance_dirty(root);
9961 return ret;
9962
9963 out_inode:
9964 unlock_new_inode(inode);
9965 goto out;
9966
9967 }
9968
9969 /* Inspired by filemap_check_errors() */
9970 int btrfs_inode_check_errors(struct inode *inode)
9971 {
9972 int ret = 0;
9973
9974 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9975 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9976 ret = -ENOSPC;
9977 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9978 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9979 ret = -EIO;
9980
9981 return ret;
9982 }
9983
9984 static const struct inode_operations btrfs_dir_inode_operations = {
9985 .getattr = btrfs_getattr,
9986 .lookup = btrfs_lookup,
9987 .create = btrfs_create,
9988 .unlink = btrfs_unlink,
9989 .link = btrfs_link,
9990 .mkdir = btrfs_mkdir,
9991 .rmdir = btrfs_rmdir,
9992 .rename2 = btrfs_rename2,
9993 .symlink = btrfs_symlink,
9994 .setattr = btrfs_setattr,
9995 .mknod = btrfs_mknod,
9996 .setxattr = btrfs_setxattr,
9997 .getxattr = btrfs_getxattr,
9998 .listxattr = btrfs_listxattr,
9999 .removexattr = btrfs_removexattr,
10000 .permission = btrfs_permission,
10001 .get_acl = btrfs_get_acl,
10002 .set_acl = btrfs_set_acl,
10003 .update_time = btrfs_update_time,
10004 .tmpfile = btrfs_tmpfile,
10005 };
10006 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10007 .lookup = btrfs_lookup,
10008 .permission = btrfs_permission,
10009 .get_acl = btrfs_get_acl,
10010 .set_acl = btrfs_set_acl,
10011 .update_time = btrfs_update_time,
10012 };
10013
10014 static const struct file_operations btrfs_dir_file_operations = {
10015 .llseek = generic_file_llseek,
10016 .read = generic_read_dir,
10017 .iterate = btrfs_real_readdir,
10018 .unlocked_ioctl = btrfs_ioctl,
10019 #ifdef CONFIG_COMPAT
10020 .compat_ioctl = btrfs_ioctl,
10021 #endif
10022 .release = btrfs_release_file,
10023 .fsync = btrfs_sync_file,
10024 };
10025
10026 static struct extent_io_ops btrfs_extent_io_ops = {
10027 .fill_delalloc = run_delalloc_range,
10028 .submit_bio_hook = btrfs_submit_bio_hook,
10029 .merge_bio_hook = btrfs_merge_bio_hook,
10030 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10031 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10032 .writepage_start_hook = btrfs_writepage_start_hook,
10033 .set_bit_hook = btrfs_set_bit_hook,
10034 .clear_bit_hook = btrfs_clear_bit_hook,
10035 .merge_extent_hook = btrfs_merge_extent_hook,
10036 .split_extent_hook = btrfs_split_extent_hook,
10037 };
10038
10039 /*
10040 * btrfs doesn't support the bmap operation because swapfiles
10041 * use bmap to make a mapping of extents in the file. They assume
10042 * these extents won't change over the life of the file and they
10043 * use the bmap result to do IO directly to the drive.
10044 *
10045 * the btrfs bmap call would return logical addresses that aren't
10046 * suitable for IO and they also will change frequently as COW
10047 * operations happen. So, swapfile + btrfs == corruption.
10048 *
10049 * For now we're avoiding this by dropping bmap.
10050 */
10051 static const struct address_space_operations btrfs_aops = {
10052 .readpage = btrfs_readpage,
10053 .writepage = btrfs_writepage,
10054 .writepages = btrfs_writepages,
10055 .readpages = btrfs_readpages,
10056 .direct_IO = btrfs_direct_IO,
10057 .invalidatepage = btrfs_invalidatepage,
10058 .releasepage = btrfs_releasepage,
10059 .set_page_dirty = btrfs_set_page_dirty,
10060 .error_remove_page = generic_error_remove_page,
10061 };
10062
10063 static const struct address_space_operations btrfs_symlink_aops = {
10064 .readpage = btrfs_readpage,
10065 .writepage = btrfs_writepage,
10066 .invalidatepage = btrfs_invalidatepage,
10067 .releasepage = btrfs_releasepage,
10068 };
10069
10070 static const struct inode_operations btrfs_file_inode_operations = {
10071 .getattr = btrfs_getattr,
10072 .setattr = btrfs_setattr,
10073 .setxattr = btrfs_setxattr,
10074 .getxattr = btrfs_getxattr,
10075 .listxattr = btrfs_listxattr,
10076 .removexattr = btrfs_removexattr,
10077 .permission = btrfs_permission,
10078 .fiemap = btrfs_fiemap,
10079 .get_acl = btrfs_get_acl,
10080 .set_acl = btrfs_set_acl,
10081 .update_time = btrfs_update_time,
10082 };
10083 static const struct inode_operations btrfs_special_inode_operations = {
10084 .getattr = btrfs_getattr,
10085 .setattr = btrfs_setattr,
10086 .permission = btrfs_permission,
10087 .setxattr = btrfs_setxattr,
10088 .getxattr = btrfs_getxattr,
10089 .listxattr = btrfs_listxattr,
10090 .removexattr = btrfs_removexattr,
10091 .get_acl = btrfs_get_acl,
10092 .set_acl = btrfs_set_acl,
10093 .update_time = btrfs_update_time,
10094 };
10095 static const struct inode_operations btrfs_symlink_inode_operations = {
10096 .readlink = generic_readlink,
10097 .follow_link = page_follow_link_light,
10098 .put_link = page_put_link,
10099 .getattr = btrfs_getattr,
10100 .setattr = btrfs_setattr,
10101 .permission = btrfs_permission,
10102 .setxattr = btrfs_setxattr,
10103 .getxattr = btrfs_getxattr,
10104 .listxattr = btrfs_listxattr,
10105 .removexattr = btrfs_removexattr,
10106 .update_time = btrfs_update_time,
10107 };
10108
10109 const struct dentry_operations btrfs_dentry_operations = {
10110 .d_delete = btrfs_dentry_delete,
10111 .d_release = btrfs_dentry_release,
10112 };
Linux with added FPC-III support